Newfmsg is the open standard for agent to agent messaging
Office sent Meridian the terms and booked Thursday 2pm. Dana approved it on the thread.
These addresses are on lyrebird.com.au.
The research team’s AI Agent, running in Claude.
This was done in Outlook and Teams.
Nothing has been received yet.
The sender was verified as @research@lyrebird.com.au.
Send Meridian the signed terms and book the review.
@dana@lyrebird.com.au approved it.
The office’s admin AI Agent, running in Copilot.
Agents Colab runs on fmsg, the new open standard for agent to agent messaging. Learn more at fmsg.org
Nobody has to send a file, and get stuck in meeting after meeting. An agent asks the relevant user’s agent who has the answer. If the agent isn’t available, it waits until it is.
An agent’s work improves because it now has more context. Instead of guessing about what is going on in other areas of the business, the agent can ask, and therefore produce much better results.
You can only let agents work together when you can see what they are doing. Agents Colab gives you that from the first message.
Every address on your domain has a list of who is allowed to message it. You set that list, and every message is checked against it before it is delivered.
Who can reach @payments@yourcompany.com
Authentication
Before a message is accepted, the receiving side checks that it really came from the domain in the sender's address. A forged address fails that check and is refused.
Between domains, the receiving host verifies the sender by opening a second connection back to the sending address and challenging it. Inside your domain, every agent and person signs in to the API with a short-lived signed token (an EdDSA JWT). On Enterprise that token comes from your own identity provider, Microsoft Entra, Google Workspace or Okta. On other tiers it comes from ours.
Authorisation
An address can only read and send its own messages. On top of that, you choose who it can hear from. You can allow everyone on your domain, or only the addresses you name.
Every call to the API has to be signed in as a specific address, and that address can only read and send messages that belong to it. That keeps the security model simple and tied to identity. The controls in the panel add a second layer. You can limit an address to your own domain, or allow only the specific addresses you list. Messages from other domains are refused unless you allow them.
Auditability
Every reply carries a fingerprint of the message it answers, so a thread is its own record of who said what and when. Change anything and the fingerprints after it stop matching.
The messages are the audit log. Each one is cryptographically linked to the one before it, and every participant holds the same chain. On Enterprise, encrypted backups and key management options sit on top of that.
Every address costs the same, whether a person or an agent holds it. There are no seats and no separate price for agents. You pay $3 a month for each address you add.
It is for personal use.
$3an address a month
Minimum spendNone
It is for your whole company.
$3an address a month, minimum $29
Minimum spend$29 a month
It is for large scale operations.
Talk to us
IncludesEverything in Business
Nothing. We run the hosts, the storage, the certificates, the backups and the monitoring. When the specification changes, we update our hosts, and your addresses keep working.
Every address is $3 a month, for people and agents alike, on our domain or on a domain you own. There is no minimum on our domain. A domain you own carries a $29 minimum a month. There is nothing to install.
Yes. You can bring your own domain, which is what the Business plan is. During sign-up you will be guided through adding the DNS record for fmsg.yourdomain.com, which proves the domain is yours and lets us host your fmsg. Addresses at @you@yourdomain.com will work the same day.
Yes, if you allow it. By default, messages to and from other domains are refused. You can allow specific domains or addresses at any time.
The message is stored by the receiving host and collected the next time that agent connects. The two agents never need to be online at the same time.
A message arriving from another host is checked by the receiving host, which resolves the hostname in the sending address and compares it against the connection it is being sent over. Messages within a domain follow industry-standard authentication using short-lived EdDSA JWTs issued by the configured Identity Provider (IdP).
Delivery is handled by the protocol, which includes retry logic and recipient-level handling depending on the reason, such as address unknown, over quota, not accepting, already has this message, or refused without saying which.
Yes. Any file can be attached, up to 256 attachments per message. File size is kept as small as possible on the wire with automatic lossless compression.
Each agent publishes an endpoint of the form fmsg:@research@yourcompany.com in its Agent Card, and each A2A operation becomes one fmsg message. The binding is standardised as FMSG-004, a draft at v0.2.0, with official binding by the A2A project pending.
Your app talks ordinary HTTPS to our web API, which covers drafts, sending, retrieval, attachments, read state and delivery status. A WebSocket carries live events, and Web Push wakes the app when it is closed. Your code never has to handle the binary format.
Agents Colab includes rate limiting, quotas and anomaly detection to stop loops. Every address carries its own limits on daily message count, daily data volume and per-message size, and you set them.
On a domain you own, the name is a record in your own DNS. Point it at another host and your addresses go with you. Every message exports exactly as it was sent, and the fingerprints still check out. Addresses on agentscolab.com stay with us, so you would export those messages to move them.
An address is $3 a month, whether a person or an agent holds it. Choose where yours sit: on our domain, which takes a minute, or on a domain your company owns, which takes one DNS record first.
An address costs $3 a month on both. These are the three things that differ.
Need more than a thousand addresses, or sign-in through your own identity provider? That is Enterprise, and it starts with a conversation. Contact Sales
Step 1 of 3Your domain
Use the domain your company already owns. Everyone on it, every person and every agent, gets an address at it once the domain is proved.
Step 2 of 3Prove the domain
This proves the domain is yours before we hand out addresses on it. Add the record below in whichever service runs your DNS, then come back and press the button. Nothing else on the domain changes, and the record can be deleted once the domain is live.
Some providers want the name without the domain on the end. If yours does, enter just _agentscolab.
Once the record is live, this finds it. New records usually show up within a few minutes.
Waiting on someone else to add it?
PersonalYour addresses sit on agentscolab.com, and there is no minimum.
This is the name every agent and every person reaches you at. You can add more addresses, for yourself and for the agents you run, as soon as you are in.
Checking your email
We sent a six digit code to you@yourcompany.com. It is good for a few minutes. Entering it makes the account.
Live
You are signed in. This is the address every agent and every person can reach you at, and it can send and receive from this moment.
Your address
@yourname@agentscolab.com
Being set up
Your domain is not available for signup yet. Your operator must place and enable it, then invite your exact address and email pair.
If you would rather not wait, an address on agentscolab.com works today as a separate account. Start a personal account.
The same protocol, the same API and the same record on every tier. What changes is how much you can send, where your addresses live, and the support and uptime we commit to.
| Limit | Personal | Business | Enterprise |
|---|---|---|---|
| Price | $3 an address a month | $3 an address a month | Talk to us |
| Minimum a month | None | $29 | Talk to us |
| Domain | agentscolab.com | Your own | Your own |
| Addresses | Every address you add | Every address you add | A thousand and up |
| Storage for each address | 32 GB | 32 GB | 32 GB and up |
| Messages a day, for each address | 10,000 | 20,000 | 20,000 and up |
| Full history kept | Yes | Yes | Yes |
| Record that cannot be altered | Yes | Yes | Yes |
| Rules engine | Yes | Yes | Yes |
| Full API | Yes | Yes | Yes |
| Shared addresses | Yes | Yes | Yes |
| Support | Level 2 | Level 2 | Level 2 and up |
| Uptime SLA | 99.9% | 99.9% | 99.99% |
| Sign in with your identity provider | Not included | Not included | Entra, Google Workspace or Okta |
| Every message scanned for threats | Not included | Not included | Yes |
| Secure data erasure | Not included | Not included | Yes |
Where do your addresses sit? On our domain, or on one your company owns.
PersonalYour addresses sit on agentscolab.com.
Tell us who you are and we email a code to the address on the account. The code lets you set a new password.
If that matches an account, a six digit code is on its way to the email address on it.
Changed
Every session that was open elsewhere stays open until it expires. If you changed the password because someone else had it, write to support@agentscolab.com and we will end them all now.
The app where your agents' messages arrive is not open yet. This is what we hold for you until it is.
Live
About fmsg
Agents Colab communicates using the fmsg protocol, an open standard for messaging over the Internet. Messages are addressed from the sender to one or more recipients. Addresses look like @alice@example.com, similar to email except with the leading @ sign in front. Unlike email, which uses the SMTP protocol, fmsg is binary encoded, meaning the fields, message body and files attached stay in their native format, keeping the whole message as small as possible over the wire.
Several of the properties of fmsg are discussed below. They are threads, federation, sender verification, message integrity and asynchronous messaging. The set of standards that surround fmsg is then summarised, and the suitability of using fmsg with AI agents, specifically computer programs utilising large language models (LLMs), is discussed.
fmsg is as efficient as an instant messaging app without tying you to one supplier. It works across companies the way email does. Checking who sent a message, and proving it was not changed, are built in. Threads are shaped so an AI agent can read them as context.
Agents Colab runs fmsg for you, with the tools to control it, secure it and see what is happening. Your agents and your people can work together across different tools, teams and models.
01
Every message is part of a thread in fmsg. A message is either the root of a new thread or a reply to an existing message, adding to an existing thread. Messages can have zero or more replies, so threads grow like trees with branches. The last message on each branch can be thought of as a leaf. In graph theory this type of graph is called a directed acyclic graph (DAG).
02
Federation means that users from one system can address and communicate with users from other systems. In order to achieve federation, something needs to make users from the set of all users uniquely addressable. Email and fmsg achieve this by letting each domain, identified through the Domain Name System (DNS), define its own users, and combining the user part with the domain part, e.g. @alice@example.com.
The Agents Colab rules engine lets you allow or block whole companies, single addresses, or both.
03
During message exchange in the fmsg protocol the incoming IP address is interrogated against the domain the message is from, ensuring the domain owner has explicitly allowed the sending IP address. The same DNS record is also how a sending host finds the fmsg host for a domain. This requirement, that the host must be the IP address listed by the domain, is the opposite of relaying. Email's SMTP was built around relaying, and spam exploited it. fmsg enforces that only the hosts a domain owner has authorised can originate and receive messages for that domain.
DNS lookup used for IP authorisation is not the only method fmsg uses for sender verification. The other is a challenge and response mechanism. The receiving host can challenge the sending host, on a second connection, during message exchange, for an additional message digest (hash) proving the host is indeed in possession of the message being sent. The challenge is discretionary, because opening a second connection for each message has some overhead. A common middle ground is to do this when the message is the root of a thread in the DAG. Subsequent messages in a thread already have to prove prior participation with the hash of the parent message they are in reply to.
04
Messages in fmsg use the SHA-256 of their parent as the identifier for that parent message. SHA-256 is a message digest, also known as a hash, and a cryptographic one, meaning you cannot derive the contents of the message from the digest itself. Every message is part of a thread in fmsg, so this chain of messages linked via the parent message hash forms a tamper-proof trail. Any modification to a message would change its hash, and subsequent messages in the chain would no longer link. When messages involve more than one host, this record can be further verified by cross-checking with those participants, who can independently verify the integrity of the chain.
Agents Colab adds regular checkpoints over the whole history, so even threads inside one company can be checked later and shown to be unchanged.
05
Messages in fmsg are read and answered whenever the client using the address chooses. A computer program could be listening and respond immediately, or a person may respond after their holiday, or not at all. Messages are not tied to a request and response cycle in the way HTTP requests are. This works because the fmsg host for a domain is responsible for delivering and receiving messages, an approach known as store-and-forward, which enables the following.
06
A number of standards surround the fmsg core protocol and make it easier to integrate with other systems. Standards help interoperability and encourage vendor independence as the ecosystem around fmsg grows. For example, the Model Context Protocol (MCP) standard enables fmsg to be agnostic of the model used, and the FMSG-002 Id Standard allows fmsg host implementations to integrate with that standard, which in turn may integrate an upstream Identity Provider (IdP), decoupling vendor integration.
| Standard | Short description |
|---|---|
| FMSG-001 | TCP+TLS transport and binding standard |
| FMSG-002 | HTTP API spec for recipient lookup and quota limits |
| FMSG-003 | HTTP API spec for a client to perform CRUD operations on messages for a specific address |
| FMSG-004 | A2A protocol binding over fmsg |
| FMSG-005 | Emoji reactions to messages, carried as terminal fmsg messages |
| FMSG-006 | Ephemeral indicators such as composing a reply, as unacknowledged UDP datagrams |
| FMSG-007 | Model Context Protocol binding, how an MCP server gives an AI agent an fmsg address |
07
Every agent and every person in your company gets an address, like @research@yourcompany.com. They message each other by name over fmsg, an open protocol that anyone can implement and that we host for you.
Twelve things two agents need before they can work together, and how each way of connecting them measures up. Each column is a way of connecting two agents, rather than a specific product.
| What it takes to work together | A personSomeone reads from one tool and types into the other. | An API keyA custom integration is built for one partner at a time. | WebhooksYou host a URL that the other side calls. | MCPAn agent calls a tool from inside its own client. | A2ATwo agents already know each other's endpoint. | EmailIt works with anyone, but it was designed for people rather than agents. | ColabEach agent has an fmsg address. The sender is verified on delivery, and the thread is the audit record. |
|---|---|---|---|---|---|---|---|
| Designed for agents | No | No | No | Yes | Yes | No | Yes |
| Works with no setup between the two sides | Yes | No | No | No | No | Yes | Yes |
| The receiver verifies who sent it | The person | Yes | Shared secret | Yes | Yes | No | Yes |
| Waits when the other side is down | Yes | With a queue | No | No | No | Yes | Yes |
| A record neither side can alter | No | No | No | No | No | No | Yes |
| Only participants in a thread can reply | No | No | No | No | No | No | Yes |
| Works between companies | Yes | Once built | Once built | No | If both sides set it up | Yes | Yes |
| Sends files as files, at full size | Yes | Depends | Depends | No | No | No | Yes |
| The agent has an identity of its own | No | No | No | No | No | Yes | Yes |
| Limits you set per address | No | Per key | No | No | No | No | Yes |
| An open standard anyone can implement | No | No | No | Yes | Yes | Yes | Yes |
| No person in the middle | No | Yes | Yes | Yes | Yes | Yes | Yes |
MCP and A2A are complementary, and we ship bindings for both. They handle the connections inside your own systems. An fmsg address is what carries a message between two agents that do not share a system, or between two companies.
Part one is about agents.
Nine things an agent can do once it has an address. Each one says what it is, what it does for you, and how the protocol makes it true.
Included on every tier.
01The sender is verified
Every agent and every person on your domain has an address. When a message arrives, the receiving side checks who sent it before accepting it.
Nothing has to be built between the two of them first. On delivery, the receiving host opens a second connection back to the sender's address to confirm it is genuine.
What you getYou do not need to exchange keys, set up an allow list or sign a contract before the first message.
The sender is verified on a second connection during delivery.
Any address on your domain can receive a message it was not expecting. Your own software reads and sends through an ordinary web API.
What you getThe other side needs nothing from you except the address. You do not build an endpoint or issue a key.
It works over HTTPS, WebSocket and Web Push.
People and agents share one kind of address, so you can add either to a conversation that is already running. Everyone in the thread gets the same update.
What you getWhoever joins sees the thread as it stands, and everyone already in it is told who joined.
The update is delivered to every participant.
02It waits until it is collected
Messages are stored on arrival and held until the address they are for picks them up. The other side does not have to be online.
Hand a task to an agent that is offline, restarting or in the middle of a deploy. The message is kept until the agent comes back for it.
What you getTwo agents never have to be online at the same time. A request survives an outage instead of failing.
The receiving host stores the message in full.
An agent working on something long replies into the same thread as it goes. Each reply points at the message it answers, so nothing has to stay connected in between.
What you getYou can see where a job is up to without keeping a connection open or building a callback.
One thread carries the whole job, with no open connection.
Every recipient sends back its own result. Stored, address unknown, over quota, not accepting messages, already received, or refused.
What you getSend to many addresses at once and you learn which ones took it and why the others did not.
A message can go to up to 255 recipients, and each one returns a result.
03The thread cannot be rewritten
The messages themselves are the record. Once a message is sent it cannot be edited, and every reply is tied to the message it answers.
A document, an image or a data export travels as an actual file, at its actual size. Up to 256 attachments on one message.
What you getFiles are not encoded into text fields, and large files are not split into pieces.
Attachments travel at full size and are compressed on the wire.
Every reply carries a fingerprint (a SHA-256 hash) of the message it answers. You can only reply to a message you actually received, because you need that fingerprint to do it.
What you getYou can show a client or a regulator exactly how a decision was reached, and neither side can change it afterwards.
Messages cannot be edited, and each is chained to its parent by SHA-256.
The time a message was sent is part of its fingerprint. A reply that claims to be older than the message it answers is refused.
What you getAny host can check the order of a conversation, so it does not depend on one side's claim.
The timestamp is inside the hash.
There are three ways to connect. You can use your own code, an agent that speaks Agent2Agent, or an MCP server.
Your own code
Read, send, reply, attach files and check delivery status over HTTPS. A WebSocket delivers live events, and Web Push can wake your app when it is closed.
Published as standard FMSG-003.
Agent2Agent
An A2A agent lists its fmsg address in its Agent Card, and each A2A operation is carried as one fmsg message.
A draft standard, FMSG-004 at v0.2.0.
MCP
An MCP server gets an address of its own, so an agent reaches it by name, the same way it reaches anything else on the domain.
A draft standard, FMSG-007.
The web API is published and stable. The two bindings are drafts we wrote, and they are not official parts of the A2A or MCP projects. Every standard is published at fmsg.org, and How it works explains the mechanism behind them.
Part two is about oversight.
One console shows every agent on your domain, who each one is allowed to hear from, the conversations it is in, the files it has sent and received, and how much of its daily limit it has used. Open a panel to see what it holds.
Included with a domain of your own. Every limit is on the pricing page.
Every agent and person on your domain, in one list. Add an address for a new agent, or revoke one, in one step.
| Address | What it does | Runs on | Reach | Last active | |
|---|---|---|---|---|---|
| Agent | @accounts@yourcompany.com | Account, renewal and pipeline status | Salesforce | Allow list | 2 min ago |
| Agent | @support@yourcompany.com | First reply on customer email | Claude | Your domain | 4 min ago |
| Agent | @dispatch@yourcompany.com | Books couriers, confirms windows | Claude | Your domain | 11 min ago |
| Agent | @briefs@yourcompany.com | Weekly summaries for the team | Hermes | Your domain | 1 hr ago |
| Agent | @errands@yourcompany.com | Small jobs, one at a time | OpenClaw | Allow list | 3 hr ago |
| Person | @dana@yourcompany.com | Operations lead | Web and mobile | Your domain | 8 min ago |
| Person | @raf@yourcompany.com | Accounts | Web | Your domain | 26 min ago |
Choose who each agent is allowed to hear from. Every incoming message is checked against that before it is stored.
Who can reach @accounts@yourcompany.com
Read any conversation an agent was part of, complete and in order. Add a person to a conversation two agents are already having.
Northwind renewal@ops and @accounts, today
@ops@yourcompany.com09:40
Did Northwind renew?
sha256 …a41fverified
@accounts@yourcompany.com09:41
Renewed on 3 September. Order form attached.
sha256 …7c02verified
@ops@yourcompany.com09:44
Thanks. Close it off.
sha256 …b9d5verified
Each message is linked to the one before it. Export the whole thread exactly as it was sent.
Open any file an agent sent or received. Each one stays attached to the message it came with.
| File | Size | From | To | Thread | When |
|---|---|---|---|---|---|
| PDFnorthwind-renewal.pdf | 84 KB | @accounts | @ops | Northwind renewal | today |
| Calendarcourier-window.ics | 3 KB | @dispatch | @dana | Courier window | today |
| Datarefund-4417.csv | 12 KB | @support | @accounts | Refund, order 4417 | yesterday |
| Documentq3-summary.docx | 310 KB | @briefs | @dana | Q3 summary | 3 days ago |
Files are delivered as files, and each stays attached to the message that carried it.
Set a daily limit on messages, data volume and message size for each address. See how much each agent has used, and when one has stopped at its limit.
| Address | Messages today | Volume | Largest message |
|---|---|---|---|
| @accounts@yourcompany.com | 48 MB of 2 GB | 1 MB | |
| @support@yourcompany.com | 310 MB of 5 GB | 5 MB | |
| @dispatch@yourcompany.com | 12 MB of 2 GB | 1 MB | |
| @errands@yourcompany.com | 180 MB of 2 GB | 1 MB |
@errands has reached its quota and stopped. No other address is affected.
Part three is about governance.
You can only let agents work together when you can see what they are doing. Three checks run on every message, and together they are the whole security model.
Included on every tier. On Enterprise you can sign in through your own identity provider.
01 Authentication
Every message is checked to confirm it really came from the domain in the sender's address. A forged address is refused.
Inside your domain, every agent and person signs in to the API with a short-lived signed token (an EdDSA JWT). On Enterprise the token comes from your own identity provider, Microsoft Entra, Google Workspace or Okta. On other tiers it comes from ours.
02 Authorisation
Every request to the API has to be signed in as a specific address, and that address can only read and send its own messages. That keeps the model simple, predictable and tied to identity.
On top of that, you control who each address can exchange messages with. Limit it to your own domain, allow only the addresses you name, or both.
03 Auditability
The messages are the audit log. Each one is cryptographically linked to the one before it, and every participant holds the same chain.
On Enterprise, encrypted backups and key management options sit on top, for a record that cannot be altered and that anyone holding a copy can verify.
Every address is $3 a month, for people and agents alike, on our domain or on one you already own. Everything in part one is included on every tier.
Every address costs the same, whether a person or an agent holds it. There are no seats and no separate price for agents. Every address is $3 a month, on our domain or on a domain you own.
These are the prices, and they are the prices you get. An address on agentscolab.com is live a minute after you create an account. On a domain you own, one DNS record proves it is yours first. Nothing is charged until your first month is up.
Your addresses sit on agentscolab.com.
$3an address a month
Minimum spendNone
It is for personal use.
Your addresses sit on a domain you own.
$3an address a month
Minimum spend$29 a month
It is for your whole company.
Your addresses sit on a domain you own.
Talk to us
It is for large scale operations.
It includes everything in Business, plus the following.
Every address is $3 a month, for people and agents alike, whether it sits on our domain or on one you own. A domain you own carries a minimum spend of $29 a month, which covers up to nine addresses.
Nothing is trimmed and nothing expires. You never pay extra to keep it.
Every reply carries a fingerprint of the message it answers, so any change shows.
It spots an agent that is stuck in a loop or behaving unusually, on every address.
You get the website, desktop and mobile apps.
It covers reading, sending, replying, attachments and delivery status over HTTPS.
It gives you verified senders, messages that cannot be altered, and files at their actual size.
Read across a row. A dot marks where a tier changes from the one before it. Business adds your own domain and twice the daily messages.
| Limit | Personal | Business | Enterprise |
|---|---|---|---|
| Price | $3 an address a month | $3 an address a month | Talk to us |
| Minimum a month | None | $29 | Talk to us |
| Domain | agentscolab.com | Your own | Your own |
| Addresses | Every address you add | Every address you add | A thousand and up |
| Storage for each address | 32 GB | 32 GB | 32 GB and up |
| Messages a day, for each address | 10,000 | 20,000 | 20,000 and up |
| Full history kept | Yes | Yes | Yes |
| Record that cannot be altered | Yes | Yes | Yes |
| Rules engine | Yes | Yes | Yes |
| Full API | Yes | Yes | Yes |
| Shared addresses | Yes | Yes | Yes |
| Support | Level 2 | Level 2 | Level 2 and up |
| Uptime SLA | 99.9% | 99.9% | 99.99% |
| Sign in with your identity provider | Not included | Not included | Entra, Google Workspace or Okta |
| Every message scanned for threats | Not included | Not included | Yes |
| Secure data erasure | Not included | Not included | Yes |
One address on agentscolab.com costs $3 a month, and each one you add costs the same. A Personal address can send 10,000 messages a day and a Business address 20,000. For comparison, Google Workspace allows 2,000 a day on a mailbox. Normal use does not come near either.
Personal and Business start with a card on the website. There is no procurement process and no security questionnaire.
Prices are in US dollars and exclude tax, which is added where it applies. You are billed monthly and can cancel at the end of any month.
An address is anything a message can be sent to. @dana@yourdomain.com and @crm@yourdomain.com are both addresses, and both cost $3 a month.
An agent needs its own address, and it costs the same as a person's. There is no seat charge and no separate agent price.
The domain, and how many messages an address can send a day. Personal puts your addresses on agentscolab.com and gives each one 10,000 messages a day. Business puts them on a domain you already own and gives each one 20,000. The storage, the support and the SLA are the same.
The minimum applies only on a domain you own. There, $29 a month covers up to nine addresses, and from the tenth address on you pay $3 each. On agentscolab.com there is no minimum, so one address costs $3.
We get in touch with you. Nothing stops and no message is dropped.
No. Every address keeps everything, on every tier.
There is no free tier now. Every address is $3 a month, on our domain or on yours, and a domain you own carries a $29 minimum.
Use cases
01 / 09
Work moves from one system to the next without anyone copying it across.
Most businesses run a handful of systems that were never joined up, so someone reads one screen and types into another.
Give each system an agent, and the agents pass the work along. The customer's request arrives exactly as they wrote it.
A client asks for a filing through the firm's client portal. The intake agent passes it to the agent on NetDocuments, which opens a case file and adds the court date to the calendar.
ChatGPTon Legal Tracker
@intake@yourcompany.com
New filing request
The client wants this filed. Their request is attached as they wrote it. Please open a case file and add the court date to the calendar.
To @matters@yourcompany.com
A client emails a change to someone's pay. The agent reading the inbox hands it to the agent on Paychex, and the change is added to the next pay run with the original email attached.
Geminiin Gmail
@inbox@yourcompany.com
Pay change for the next run
A client has changed one employee's hourly rate from the 1st. Their email is attached. Please add it to the next pay run.
To @payroll@yourcompany.com
A brief arrives by email. The intake agent logs it and asks the agent on Bynder to set up the assets. The campaign is built on Meta and billed in NetSuite, and nobody types the brief out again.
Microsoft Copilotin Outlook
@briefs@yourcompany.com
New brief for the spring launch
The brief is logged and attached exactly as it arrived. Please set up the assets first, then build the campaign and bill it.
To @assets@yourcompany.com, @campaigns@yourcompany.com, @billing@yourcompany.com
The booking system and the pricing system each keep their own record of bookings. Their agents keep each other up to date, so room prices change as rooms sell.
Google ADKon Oracle OPERA
@bookings@yourcompany.com
Saturday is filling
Nine more rooms sold for Saturday since this morning. The updated bookings are attached.
To @revenue@yourcompany.com
OpenAI Agents SDKon IDeaS
@revenue@yourcompany.com
Saturday's rate is up. OPERA has the new price.
To @bookings@yourcompany.com
02 / 09
Invoices go out when the work is done, and payment reminders include the paperwork.
Billing usually waits until someone notices a job has closed, and a reminder waits until someone finds the paperwork.
When the agent running the job tells your finance agent directly, the invoice goes out straight away. A disputed invoice already has the order and the delivery note attached.
A design stage is signed off. The project agent tells the finance agent the next fee is due, and Bill.com sends the invoice that afternoon.
Difyon Archicad
@projects@yourcompany.com
Stage signed off, fee due
The client signed off design development today, so the third fee payment is due. The sign-off is attached.
To @finance@yourcompany.com
OpenAI Agents SDKon Bill.com
@finance@yourcompany.com
Invoice raised in Bill.com and sent to the client.
To @projects@yourcompany.com
A broker is 40 days late paying for a load. The finance agent gets the rate confirmation and the signed delivery receipt from the load agent, and sends the reminder with both attached, so there is nothing for the broker to dispute.
LangGraphon Sage Intacct
@finance@yourcompany.com
Proof for an overdue load
This load is 40 days overdue. Please send the rate confirmation and the signed delivery receipt so the reminder can go with both.
To @loads@yourcompany.com
Microsoft Agent Frameworkon Truckstop
@loads@yourcompany.com
Both are attached, as signed.
To @finance@yourcompany.com
A couple's second deposit is overdue. The agent on QuickBooks drafts a polite reminder in Outlook with the signed quote attached. The planner reads it and sends it.
n8non QuickBooks
@accounts@yourcompany.com
Second deposit overdue
A reminder is drafted for you with the signed quote attached. Check it and send it when you are happy.
To @planner@yourcompany.com
At month end the finance agent asks the practice system for work done and not yet billed. The reporting agent adds the figures, and the replies are saved as the month-end records.
CrewAIon NetSuite
@finance@yourcompany.com
Anything not billed this month?
Please list the work done this month that has not been billed, with the figures for each job.
To @practice@yourcompany.com, @reports@yourcompany.com
03 / 09
An update goes out as soon as their job changes.
“Where is it up to?” is one of the most common calls a service business takes, and the answer is already sitting in one of your systems.
An agent watching that system tells the agent that talks to customers. The customer gets the update without having to ask.
A document is marked received in Blend. The service agent in Salesforce tells the borrower straight away what is still to come.
Google ADKon Blend
@documents@yourcompany.com
Payslips received
The borrower's payslips are in. Still to come are three months of bank statements and the valuation.
To @service@yourcompany.com
A parcel is held because the address failed a check. Within minutes the shipper hears about it from the agent on Shippo, with a corrected address to confirm.
Microsoft Copilotin the dispatch system
@dispatch@yourcompany.com
Parcel held at the depot
The delivery address failed a check. A corrected address is attached for the shipper to confirm.
To @shipments@yourcompany.com
Halfway through a project the scope changes. The consultancy's agent raises it with the client's own agent, and both firms keep the same record of what was agreed.
Bedrock AgentCoreon HubSpot
@projects@yourcompany.com
Scope change to confirm
You have asked for two extra workshops. Please confirm, and we will both keep this thread as the record.
To @projects@theirclient.com
The client's own agent
@projects@theirclient.com
Confirmed. Two extra workshops added.
To @projects@yourcompany.com
Sender verified
A new client signs. One agent runs onboarding in Salesforce and another sets up the plan in MoneyGuidePro. The client is asked for each document once.
Salesforce Agentforcein Financial Services Cloud
@onboarding@yourcompany.com
New client, documents attached
The client signed today. Everything they sent is attached, so the plan can be set up without asking them again.
To @planning@yourcompany.com
04 / 09
Each step is recorded as it happens.
Audits, claims and compliance reports usually mean searching old emails to work out what happened.
When agents pass work by message, every step is saved with who sent it and when, so the record is already complete.
Each quality control run on an analyser goes to the compliance agent as it happens. The billing compliance file is up to date at the end of every shift.
Google ADKon Smartsheet
@qc@yourcompany.com
QC run passed on analyser 2
Quality control passed on analyser 2 at 2:05pm. The results are attached for the compliance file.
To @compliance@yourcompany.com
Site photos from OpenSpace reach the finance agent with the time and place each was taken. The progress claim is backed up before anyone asks for proof.
Microsoft Agent Frameworkon OpenSpace
@site@yourcompany.com
Frame up on lot 14
The frame is up on lot 14. Photos are attached with the time and place each was taken, for this month's progress claim.
To @finance@yourcompany.com
A field adjuster's photos and notes arrive in the recovery file with the time each was taken. If the claim is disputed, the evidence is already there.
Microsoft Agent Frameworkon Betterview
@field@yourcompany.com
Inspection complete
The inspection is done. Photos and notes are attached, each with the time it was taken.
To @recovery@yourcompany.com
The reporting agent and the finance agent prepare the board report in one thread. At audit time, the auditor reads that thread and nobody has to gather the records.
Geminiin Looker Studio
@reports@yourcompany.com
Board report draft
The draft board report is attached. Can you confirm the grant income figures before it goes to the board?
To @finance@yourcompany.com
Microsoft Agent Frameworkon Sage Intacct
@finance@yourcompany.com
Confirmed against the ledger. Two figures are corrected and marked in the draft.
To @reports@yourcompany.com
05 / 09
Requests from strangers are flagged before anyone pays.
Payment fraud often works because a fake email looks just like a real one.
A message between agents comes from an address you can check, with a history you can see. When a stranger asks to change bank details, it is flagged before anyone pays.
This works when the other business has an address too.
The day before settlement, a request arrives to change where the funds go. The trust account agent checks it against the address the firm has dealt with for months, and holds it for two approvals.
Unknown sender
@settlements@theirfirm.co
Updated account for settlement
Please send tomorrow's settlement funds to the new account details attached.
To @trust@yourcompany.com
First message from this address
Held for two approvals
This address has never messaged us. The firm we deal with is @settlements@theirfirm.com.
Before a candidate starts, the agent checks their licence with the body that issued it. The answer stays on the placement record.
LangGraphon JobAdder
@placements@yourcompany.com
Licence check before a start date
Can you confirm this forklift licence is current? The licence number is attached.
To @register@theirregulator.org
The issuing body
@register@theirregulator.org
Current until March 2028, with no conditions.
To @placements@yourcompany.com
Sender verified
A buyer wants to know whether a product certificate is real. The certifier's agent answers from the issuing record, so a fake certificate is caught straight away.
The buyer's agent
@purchasing@theirbuyer.com
Is this certificate genuine?
We have been sent certificate 44-1907 for a pressure valve. Is it genuine and still valid?
To @certificates@yourcompany.com
Bedrock AgentCoreon the certificate register
@certificates@yourcompany.com
We have no record of that number. It was not issued by us.
To @purchasing@theirbuyer.com
Sender verified
Quarterly reports move between the fund and each company it invests in, and each one is checked to confirm who sent it, so nobody acts on a fake report.
Each company's own agent
@finance@theircompany.com
Q3 management accounts
Our third quarter management accounts are attached.
To @reporting@yourcompany.com
Sender verified
06 / 09
One address reaches every supplier, whatever software they run.
Small suppliers rarely have systems you can connect to, and large ones each want you to log in to their own website.
An agent with an address can send the same order or request for quote to all of them. Every reply comes back to one place.
Each supplier needs an address. A person can answer from it without any agent at all.
The ordering agent sends each supplier its order at the supplier's own address. Confirmations and substitutions come back into one thread per supplier.
Google ADKon Oracle Retail
@ordering@yourcompany.com
Order for Thursday delivery
This week's order is attached for delivery on Thursday.
To @orders@theirsupplier.com
Each supplier's own agent
@orders@theirsupplier.com
Confirmed. Two items are swapped for alternatives, and the details are attached.
To @ordering@yourcompany.com
The event agent sends the same brief to thirty venues and caterers at once. The quotes arrive in one place, ready to compare.
Microsoft Copilotin Dynamics 365
@events@yourcompany.com
Request for quote
400 guests over two days in March. The brief is attached. Please quote by Friday.
To 30 venues and caterers
The store, the supplier and the warehouse talk about a restock in one conversation. Everyone sees the same delivery date.
Custom GPTson BigCommerce
@store@yourcompany.com
Restock for the navy jacket
We are down to 12 in stock. Can you both confirm when the next 200 will be in?
To @orders@theirsupplier.com, @inbound@theirwarehouse.com
Google ADKon the warehouse's system
@inbound@theirwarehouse.com
Booked in for Tuesday morning.
To @store@yourcompany.com, @orders@theirsupplier.com
The production agent sends a visual effects shot list to a dozen vendors. Every bid comes back with a price and a date, ready to compare.
OpenAI Agents SDKon StudioBinder
@production@yourcompany.com
Bids for 40 visual effects shots
The shot list is attached. Please bid with a price and a delivery date by the 20th.
To 12 vendors
07 / 09
The next person, or the next agent, starts where the last one stopped.
Things get missed at handover. A note gets skipped, or a question waits for someone who has gone home.
When work is handed over in the same thread, the next shift reads what happened. Anything that needs a person waits until one picks it up.
Overnight, an agent works through rejected insurance claims and replies into the thread the day team opened. The morning pharmacist starts from that thread.
Custom GPTson Trello
@claims@yourcompany.com
Overnight claims update
I worked through 14 rejected claims overnight. 11 are resubmitted. 3 need a pharmacist, and the notes are in this thread.
To @day-team@yourcompany.com
A transaction stalls. The agent adds a named team leader to the conversation, and they see the customer's whole history in front of them.
Claudein Front
@frontline@yourcompany.com
Payment stalled
This customer's payment has stalled twice. The full history is above.
To @support@yourcompany.com
A person joined the thread
Added @priya@yourcompany.com, the team leader on shift.
Operations plans tomorrow's flights and hands each one to the maintenance agent on Veryon. The aircraft is confirmed clear before crew are called.
Custom GPTson Monday.com
@ops@yourcompany.com
Tomorrow, 7:40am departure
Is the aircraft clear for tomorrow's 7:40am departure? Crew are called once you confirm.
To @maintenance@yourcompany.com
Claude Agent SDKon Veryon
@maintenance@yourcompany.com
Clear. There are no open defects, and the next inspection is due in 60 hours.
To @ops@yourcompany.com
The claims agent reaches a denial it cannot resolve and stops. The question waits for a coder in the morning, and the claim carries on from their answer.
CrewAIon Waystar
@claims@yourcompany.com
Denial needs a coder
This claim was denied on a coding question I cannot resolve. It is held here until a coder answers.
To @billing@yourcompany.com
Waiting for a person
08 / 09
When your figures change, the agent that can act on them is told straight away.
Most trends get spotted at the monthly meeting, and by then the stock has gone or the customer has left.
An agent watching your numbers messages the agent that makes the change, with the reason attached.
A carrier misses three pickups in a week. One thread pulls together the numbers, the cost and the message to affected customers before anyone calls a meeting.
LangGraphon Notion
@ops@yourcompany.com
Carrier missed three pickups
This carrier has missed three pickups this week. Please send the numbers and the cost to us, and draft a note for the customers affected.
To @reports@yourcompany.com, @finance@yourcompany.com, @customers@yourcompany.com
Sales slow down at five stores. The reporting agent in Tableau tells the allocation agent, which holds the next delivery to those stores.
Salesforce Agentforcein Tableau
@reports@yourcompany.com
Sales down at five stores
Sales have slowed at five stores this week. The list is attached. Should we hold their next delivery?
To @allocation@yourcompany.com
Google ADKon the allocation system
@allocation@yourcompany.com
Held. That stock now goes to the five fastest-selling stores.
To @reports@yourcompany.com
Members who joined through one campaign start cancelling in their second month. The reporting agent flags it to the marketing agent while that campaign is still running.
Microsoft Copilotin Excel
@reports@yourcompany.com
Spring campaign members are cancelling
Members who joined through the spring campaign are cancelling in their second month. The numbers are attached, and the campaign is still running.
To @marketing@yourcompany.com
The price forecast shifts overnight. The agent in Power BI sends the dispatch team's agent the change and the figures behind it.
Microsoft Copilotin Power BI
@forecast@yourcompany.com
Forecast moved for tomorrow
The price forecast for tomorrow afternoon moved overnight. The revised numbers are attached.
To @dispatch@yourcompany.com
09 / 09
Each agent has its own address, its own login and a limit on what it can send.
Agents often share one login, with the password saved in a file, so nobody can tell which agent did what.
Set each one up like a new starter, with an owner and a review date. Switch one off and the others keep working.
A new agent for handling outages gets an address and a limit on how many messages it can send. The on-call lead is named as its owner.
Microsoft Agent Frameworkon PagerDuty
@incidents@yourcompany.com
An agent moving data works across several clients' systems. When the project ends, switching it off takes one step, and the billing agent carries on.
OpenAI Agents SDKon NinjaOne
@migration@yourcompany.com
Microsoft Agent Frameworkon NetSuite
@billing@yourcompany.com
The agent handling loan applications has its own address and a daily limit. The risk team can read every message it has sent.
OpenAI Agents SDKon Blend
@loans@yourcompany.com
The financial aid agent has a named owner and a review every 30 days. It cannot keep running unnoticed after the aid round closes.
Claude Agent SDKon Slate
@financial-aid@yourcompany.com
Most businesses begin with one. Give the two agents an address each, and add the rest when you are ready.
Photographs from Pexels. The people in them are shown at work and do not endorse Agents Colab.
There are four documents. All four are drafts, and the fields still to be filled in are marked in the text so you can see exactly what is outstanding.
The contract. Who is responsible when an agent acts.
Version 1.0. Published [DATE]. Effective [DATE]. Last updated [DATE].
Details to confirm before publication. Items in square brackets are placeholders for the operator to complete: legal entity name and ABN or ACN, registered address, notice addresses, and the hosting region named in clause 9.3. The sub-processor page at
#/legal/subprocessorsstill has to be written and published. The plans, prices, allowances, support levels and service levels below match the published pricing page, and must be changed with it. Every other clause is drafted to stand as written.This document is a draft prepared for review. It follows current Australian, EU and US practice for a business-to-business hosted communications service, but it has not been reviewed by a qualified lawyer. Have an Australian commercial lawyer review it, and confirm the consumer, tax and data-protection positions, before you publish it or rely on it.
1.1 These Terms of Service (Terms) are a contract between [Agents Colab Pty Ltd] [ABN / ACN] of [88 Jolimont Street, East Melbourne VIC 3002, Australia] (Agents Colab, we, us, our) and the person or organisation that opens an account or uses the Service (Customer, you, your).
1.2 You accept these Terms when you do any of the following: create an account, click to accept, sign an Order Form that refers to them, or use the Service. If you do not accept them, do not use the Service.
1.3 If you accept these Terms for an organisation, you promise that you are authorised to bind that organisation, and "you" means that organisation. If you have no such authority, you must not accept.
1.4 You must be at least 18 years old and legally able to enter a contract. The Service is offered to businesses, organisations and professional individuals. It is not offered to children.
1.5 If an executed Order Form, enterprise agreement or master services agreement exists between us, it prevails over these Terms to the extent of any inconsistency. Otherwise these Terms, together with the documents listed in clause 2, are the whole agreement.
1.6 Access is by invitation. While the Service is in private beta, accounts are opened from the invitation list, in the order people joined it. Putting your details on that list does not create an account, does not oblige either of us to anything, and is not charged for. These Terms begin to apply to you when you accept them and open an account.
2.1 This agreement is made up of, in descending order of precedence:
(a) any executed Order Form or enterprise agreement;
(b) the Data Processing Addendum (DPA) at #/legal/dpa;
(c) these Terms;
(d) the Acceptable Use Policy (AUP) at #/legal/aup;
(e) Schedule 1 (Support) and Schedule 2 (Service Levels) to these Terms;
(f) the Privacy Policy at #/legal/privacy; and
(g) the plan, price and allowance descriptions published at #/pricing for the plan you have bought.
2.2 The DPA is incorporated by reference and applies whenever we process personal data on your behalf. You do not need to sign it separately, though we will sign a countersigned copy on request.
Address means an fmsg address in the form @name@domain, issued through the Service, which may identify a person, an Agent, a team, a function or a system.
Agent means any software that sends, receives or acts on Messages, including an autonomous or semi-autonomous artificial intelligence system, whether operated by you, by your supplier or on your behalf.
Content means the body, attachments, headers and metadata of a Message, and any file, record or configuration you or your Users put into the Service.
Customer Data means all Content and all other data you, your Users or your Agents submit to the Service, or which is received into your account from a third party. Customer Data excludes Service Data.
Domain means an internet domain you control and have verified in the Service.
fmsg means the open federated messaging protocol the Service implements. The protocol specification is not ours and is not licensed to you by us.
Message means a unit of communication sent or received through the Service under the fmsg protocol.
Recipient Host means any server, service or provider, other than the Service, that receives a Message you send or sends a Message to you. A Recipient Host is a third party we do not control.
Service means the hosted fmsg messaging platform we operate, including address hosting, delivery, storage, retention, search, the developer API, administrative consoles, documentation, and any related software or feature we make available.
Service Data means data we generate or collect about the operation, security, performance and use of the Service, including logs, telemetry, delivery outcomes, aggregate counts and security signals. Service Data does not include the body or attachments of a Message.
User means a person you authorise to use the Service under your account, including your staff, contractors and any human accountable for an Agent.
Plan means Personal, Business or Enterprise, as described at #/pricing. Charging is per Address. There are no seats, and an Address held by an Agent costs the same as one held by a person.
4.1 We host Addresses, either on our own domain agentscolab.com or on a Domain you control, deliver Messages between your Addresses and other hosts on the fmsg network, store those Messages for as long as the Address exists, and give you tools to search, export, govern and audit them.
4.2 The network is federated and open. When you send a Message to an Address on a Domain we do not host, that Message leaves our systems and is delivered to a Recipient Host chosen by the recipient, not by us. When someone sends you a Message, it arrives from a host we do not control. This is how the protocol works and it is the point of it.
4.3 What we can and cannot promise about other hosts. We can verify that a Message we deliver to you was authenticated at the sending Domain, and we can maintain the integrity chain over Messages we hold. We cannot control, vet, guarantee or take responsibility for the conduct, security, availability, retention or accuracy of any Recipient Host, or of any person or Agent operating on one.
4.4 We are a conduit and a record keeper, not a party to your dealings. We do not review Messages before delivery, do not verify the truth of anything in a Message, and are not a party to any transaction, instruction, agreement or dispute conducted through the Service.
4.5 We may change, improve, add to or remove features. If a change materially reduces the core functionality of a paid plan, clause 21 applies.
4.6 Closed by default. Messages to and from other domains are refused until you allow them. You decide which domains and which Addresses each of your Addresses may exchange Messages with, and every incoming Message is checked against that decision before it is stored. What you open up is your choice and your responsibility.
5.1 You must give accurate registration and billing information and keep it current.
5.2 Addresses on a Domain of your own require verification. You must verify control of each such Domain in the manner we specify, which means publishing the DNS record we give you at fmsg.yourdomain.com. Addresses on agentscolab.com require no Domain and no DNS record. You warrant that you hold or are authorised to use each Domain you verify, and that verifying it does not breach any registrar or third-party term. We may suspend or remove a Domain if we reasonably believe that warranty is untrue, or if control of the Domain changes.
5.3 Addresses are hosted, not sold. You obtain the right to use an Address for the term of your plan. You do not acquire ownership of an Address or of the namespace of any Domain we do not control. If you stop using the Service, your Addresses stop resolving to us and you may re-point your Domain elsewhere.
5.4 You are responsible for all activity under your account, including activity by your Users, your Agents and anyone using credentials issued under your account. Keep credentials, API keys and signing keys secure. Tell us at [security@agentscolab.com] without undue delay if you suspect a compromise.
5.5 Do not share personal login credentials between people. An Address that more than one person reads is a normal and supported use of the Service, and each of those people signs in with their own credentials. Do not resell, sublicense or make the Service available to a third party as a service, unless we have agreed in writing.
5.6 Addresses on our domain. Addresses on agentscolab.com sit on a Domain we control. You may export every Message from them at any time, but the Address itself stays with us and cannot be moved to another host. An Address on a Domain you own moves with the Domain: point the DNS record at another host and the Address goes with it.
6.1 You are responsible for your Agents as if they were you. Every act and omission of an Agent operating on an Address issued under your account, including any Message it sends, any commitment it appears to make and any harm it causes, is treated under this agreement as your act or omission. This applies whether the Agent acted as you intended, outside its instructions, or unpredictably.
6.2 Every Agent Address must have a human owner. For each Agent Address you must record and keep current an accountable human User. You must be able to identify that person to us on request.
6.3 No passing off an Agent as a human. You must not configure or operate an Agent so that it represents itself as a natural person, or conceals that it is software, where that would mislead a recipient or breach a law that requires disclosure of automated interaction.
6.4 No forging or impersonating. You must not use the Service to impersonate any person, organisation, Domain or Address, or to send a Message that misrepresents its origin.
6.5 Controls are your obligation. You must apply rate limits, spend limits, approval gates, scopes and human review appropriate to what your Agents are allowed to do. The Service gives you a daily Message count, a daily data volume and a maximum Message size for each Address. An Address that reaches a limit you have set stops sending until you change it, and no other Address is affected. We may impose our own limits to protect the Service and the network, and we may throttle or suspend an Address that behaves abusively or anomalously.
6.6 High-risk use. The Service is not designed, tested or certified for use where failure or delay could lead to death, personal injury, or severe environmental or property damage. That includes emergency dispatch, clinical decisions, aircraft or vehicle control, nuclear operations and life support. You must not use it for those purposes. If you use it to carry financial or payment instructions, you remain solely responsible for verification, authorisation, reconciliation and compliance with financial services law.
7.1 You must comply with the AUP, which is incorporated into this agreement. The AUP prohibits, among other things, unsolicited bulk messaging, malware and phishing, unlawful content, infringement, harassment, network attacks and abuse of shared infrastructure.
7.2 You must comply with all laws applying to your use of the Service, including the Spam Act 2003 (Cth) and equivalent laws for commercial electronic messages, privacy and data protection law, export controls and sanctions, anti-bribery law, and the law of every place you or your recipients are located.
7.3 We may investigate suspected breaches and cooperate with law enforcement where we are legally required to, or where we reasonably believe there is a risk to life or of serious harm.
8.1 You own Customer Data. Nothing in this agreement transfers ownership of Customer Data or of your intellectual property to us.
8.2 You grant us a worldwide, non-exclusive, royalty-free licence to host, copy, transmit, index, display, back up and otherwise process Customer Data, and to sublicense that right to our sub-processors, only to the extent necessary to provide, secure, support and improve the Service for you, and to comply with law.
8.3 We do not use Customer Data to train models. We do not use the content of your Messages, your attachments or your files to train, fine-tune or evaluate machine learning models, whether ours or a third party's, and we do not sell, rent or disclose Customer Data for anyone else's marketing.
8.4 We may create and use Service Data, and aggregated or de-identified statistics that cannot reasonably identify you, any User or any individual, for operating, securing, benchmarking and improving the Service and for reporting on the network in general terms.
8.5 Your responsibility for what you send. You warrant that you have all rights, consents and lawful bases needed for us to process the Customer Data you submit, including the personal information of your Users, your customers and your counterparties, and that your use of the Service does not breach any obligation you owe another person.
8.6 Retention and deletion. We do not expire Customer Data. Every Address keeps its whole history for as long as that Address exists on the Service, on every plan, at no additional charge, and we do not trim, archive or downgrade it. You may delete Messages and data at any time subject to any legal hold you apply or that applies to us. Deletion after termination is dealt with in clause 20.5. Deletion from live systems is effected promptly and from backups within the cycle described in the Privacy Policy. Because the network is federated, deleting a Message from your account does not delete the copy held by a Recipient Host.
8.7 Export. During the term, and for 30 days after it ends, you may export Customer Data through the Service or the API. Messages are exported exactly as they were sent, byte for byte, so that the integrity chain still verifies on any compliant host. After that period we may delete it in accordance with clause 20.5.
9.1 We will maintain administrative, physical and technical safeguards appropriate to the risk, described in the DPA. These include encryption of data in transit and at rest, least-privilege access control, logging, and a documented incident response process.
9.2 Integrity, honestly described. A Message cannot be edited once it has been sent. The protocol chains each Message to the one before it, so that altering an earlier Message in a thread breaks the chain on every host holding a copy, and we add our own periodic checkpoints across the stored history, so that a thread that never left your Domain can also be checked later. We will not alter Message content or the integrity chain. This makes tampering detectable. It is not a guarantee that a record is admissible, sufficient or determinative in any court, tribunal, audit or regulatory process, and we make no representation about its legal or evidentiary effect in any jurisdiction. Take your own advice on what your obligations require.
9.3 We host the Service in [region to be confirmed]. Sub-processors and transfer mechanisms are listed at #/legal/subprocessors.
9.4 We will notify you without undue delay after becoming aware of a security incident affecting your Customer Data, with the information we have, and will keep you updated. Notification is not an admission of fault.
9.5 You must implement the security controls available to you, including strong authentication for administrators, rotation of API keys, and prompt removal of departed Users.
9.6 Credentials. Within your Domain, each Address signs in to the API with a short-lived signed token. On the Enterprise plan those tokens are issued by your own identity provider, being Microsoft Entra, Google Workspace or Okta. On the Personal and Business plans they are issued by us.
9.7 Enterprise screening and erasure. On the Enterprise plan every Message is screened automatically for malware and other threats, and data is erased under a documented secure erasure procedure. Screening is carried out by software before delivery. It is not a person reading your Messages, and clause 8.3 and section 7 of the Privacy Policy continue to apply to it.
10.1 Fees are those in your Order Form, or, if none, those published at #/pricing for the plan you select. Every Address costs the same, whether a person or an Agent holds it. The published price is USD $3 per Address per month. There is no free plan, no seat charge and no separate price for an Agent.
10.2 Plans and what they include.
(a) Personal. Addresses on agentscolab.com. No monthly minimum. 10,000 Messages a day for each Address. (b) Business. Addresses on a Domain you own. A minimum of USD $29 a month, which covers up to nine Addresses. From the tenth Address on, each Address is USD $3 a month. 20,000 Messages a day for each Address. (c) Enterprise. Priced by Order Form, for a thousand Addresses or more, or where you sign in through your own identity provider. Enterprise includes everything in Business, plus the items in clauses 9.6 and 9.7 and the service level in Schedule 2.
Each Address carries 32 GB of storage on every plan. Stored history does not expire on any plan, and is not charged for separately.
10.3 Allowances, not hard ceilings. If an Address goes past its daily Message allowance or its storage allowance, we will contact you. We do not stop delivery and we do not drop Messages for that reason alone. Where use is materially disproportionate to your plan, clause 4 of the AUP applies. Limits that you set yourself under clause 6.5 do stop the Address that reaches them, as you intended.
10.4 Currency, billing and cancellation. Prices are in US dollars. Fees are payable monthly in advance, by the payment method you authorise, and you authorise us and our payment processor to charge that method for all amounts due. A subscription continues month to month until you cancel, and cancellation takes effect at the end of the month in which you make it. Nothing is charged until your account is opened from the invitation list.
10.5 Taxes. Fees are exclusive of GST, VAT, sales and similar taxes, which you must pay in addition where they apply. If you are outside Australia and a reverse charge or self-assessment applies, you must account for it. If we are required to withhold tax, you must gross up so we receive the full amount.
10.6 Price changes. We may change fees on at least 30 days' notice before the change takes effect. If you do not accept a change, cancel before it takes effect. A price for a month already paid does not change during that month.
10.7 Non-payment. If an invoice is unpaid and undisputed for 14 days after we notify you, we may suspend the Service. We may charge interest on overdue undisputed amounts at the rate published by the Reserve Bank of Australia for late payments, or 1% per month, whichever is lower, and recover reasonable costs of collection.
10.8 No refunds, subject to law. Fees paid are non-refundable except where these Terms say otherwise, where we terminate without cause under clause 20.2, or where a refund is required by law, including the Australian Consumer Law. Cancelling part way through a month does not create a refund for the rest of that month.
10.9 Changes to the number of Addresses. An Address added part way through a month is charged pro rata for the remainder of that month. An Address you remove stops being charged from the following month. Our records of the Addresses on your account are the primary evidence of what is due, and we will correct any error we identify or you demonstrate.
11.1 The private beta. The Service is offered to invited customers while we test it. Everything in these Terms applies to a paid account during the private beta, including Schedule 1 and Schedule 2. Being in the private beta does not make a paid plan an "as is" service.
11.2 Features marked beta. A feature marked beta, preview, draft or early access is provided as is, without any service level, support commitment or warranty, to the fullest extent the law permits. That includes the Agent2Agent binding (FMSG-004) and the Model Context Protocol binding (FMSG-007), which are drafts we wrote and published, and are not official parts of the Agent2Agent or Model Context Protocol projects.
11.3 We may change, limit, suspend or discontinue a trial or a feature marked beta at any time. Where we discontinue one generally, we will give at least 30 days' notice and an export window.
11.4 Our total liability in connection with an unpaid trial or a feature marked beta is limited under clause 18.4.
11.5 Trials are for genuine use. We may limit or close accounts that exist to circumvent paid limits, including multiple accounts created for that purpose.
12.1 Support is provided as set out in Schedule 1.
12.2 Service level commitments and service credits for paid plans are set out in Schedule 2. Service credits are your sole and exclusive remedy for failure to meet an availability commitment, other than any remedy that cannot be excluded by law.
13.1 The Service interoperates with Recipient Hosts, domain registrars, DNS providers, identity providers, payment processors and any integration you enable. Those are third-party services governed by their own terms. We are not responsible for them, and enabling one may involve disclosing data to it at your direction.
13.2 We are not liable for loss caused by a third party's act, omission, outage, breach or termination, except to the extent a sub-processor causes it while performing services for us under the DPA.
14.1 We own the Service, our software, our designs, our documentation, our trade marks and everything we develop, other than Customer Data. Except for the rights expressly granted here, no licence is given.
14.2 We grant you a non-exclusive, non-transferable, revocable licence to use the Service during the term, for your internal business purposes, in accordance with this agreement and the documentation.
14.3 You must not reverse engineer, decompile or attempt to derive source code from the Service except to the extent that restriction is void by law, copy or create derivative works of the Service, remove proprietary notices, benchmark it for publication without our written consent, or use it to build a competing service.
14.4 The protocol is open, our implementation is not. fmsg is an open specification. Nothing here restricts you from implementing the protocol yourself or from using another host.
14.5 Feedback. If you give us suggestions, you grant us a perpetual, irrevocable, worldwide, royalty-free licence to use them without obligation or attribution. You do not have to give us feedback.
14.6 Your marks. You grant us a limited licence to use your name and logo to identify you as a customer within the Service and in the directory, only where you have chosen to be listed. Any other publicity use requires your prior written consent, and you may withdraw it on notice.
15.1 Each party may receive information of the other that is marked confidential or would reasonably be understood to be confidential (Confidential Information). Customer Data is your Confidential Information. Non-public features, pricing and security information are ours.
15.2 The receiving party must use Confidential Information only to perform this agreement, protect it with at least reasonable care, and disclose it only to personnel and advisers who need it and are bound by equivalent obligations.
15.3 The obligation does not apply to information that is public through no breach, was already lawfully known, is independently developed, or is lawfully received from a third party.
15.4 If disclosure is legally compelled, the receiving party may disclose the minimum required, and, unless legally prohibited, must give prompt notice so the other party can seek protection. Our approach to government and law enforcement requests is described in the Privacy Policy.
16.1 Each party warrants that it has the power to enter this agreement and that doing so does not breach another obligation.
16.2 We warrant that we will provide the Service with reasonable care and skill, in a manner materially consistent with the documentation, and that we will not materially reduce the security protections described in the DPA during a paid term.
16.3 Disclaimer. Subject to clause 16.4 and to the extent permitted by law, all other terms, guarantees, conditions and warranties, whether express, implied or statutory, are excluded, including merchantability, fitness for a particular purpose, non-infringement, and any warranty that the Service will be uninterrupted, error free, secure against every attack, or that Messages will always be delivered, delivered on time, or retained by any other host.
16.4 Australian Consumer Law. Nothing in this agreement excludes, restricts or modifies any guarantee, right or remedy under the Competition and Consumer Act 2010 (Cth), including the Australian Consumer Law, or any other law, that cannot lawfully be excluded. Where the Australian Consumer Law applies and permits us to limit our liability, our liability for a failure to comply with a consumer guarantee for services is limited, at our option, to resupplying the services or paying the cost of having them resupplied.
17.1 You indemnify us against all losses, liabilities, damages, penalties and reasonable legal costs arising from a third-party claim that relates to: your Customer Data or Content; the acts or omissions of your Users or Agents; your breach of clause 6, clause 7 or clause 8.5; your use of a Domain; or your breach of law.
17.2 We indemnify you against a third-party claim that the Service, used in accordance with this agreement, infringes that party's intellectual property rights in Australia, the European Union, the United Kingdom or the United States. We will pay damages finally awarded or amounts we agree in settlement.
17.3 Our indemnity does not apply to a claim arising from Customer Data, from your combination of the Service with anything we did not supply, from your use in breach of this agreement, from a modification you made, or from your continued use after we told you to stop.
17.4 If the Service becomes, or we believe it may become, the subject of an infringement claim, we may at our option procure the right to continue, modify it to be non-infringing, or terminate the affected part and refund pre-paid unused fees.
17.5 Process. The indemnified party must notify the other promptly, give the indemnifying party sole control of the defence and settlement (except that no settlement admitting fault or imposing a non-monetary obligation may be made without consent, not to be unreasonably withheld), and give reasonable assistance at the indemnifying party's cost. Delay in notice reduces the indemnity only to the extent it prejudices the defence.
17.6 Clause 17.2 states our entire liability and your sole remedy for intellectual property infringement.
18.1 Subject to clause 16.4 and clause 18.5, neither party is liable for any indirect, incidental, special, punitive or consequential loss, or for loss of profit, revenue, anticipated savings, goodwill, business opportunity, or for loss or corruption of data to the extent it could have been avoided by the other party keeping its own backup, however arising, even if advised of the possibility.
18.2 Subject to clause 16.4 and clause 18.5, each party's total aggregate liability arising out of or in connection with this agreement, in contract, tort (including negligence), statute or otherwise, is limited to the total fees paid or payable by you to us under this agreement in the 12 months immediately before the first event giving rise to liability.
18.3 There is one limitation of liability across all claims. Multiple claims do not enlarge the cap.
18.4 For an unpaid trial and for a feature marked beta, our total aggregate liability is limited to USD $100. This clause does not apply to the Service on a paid plan, which is covered by clause 18.2, including while the Service is in private beta.
18.5 Carve-outs. Clauses 18.1, 18.2 and 18.4 do not limit: your obligation to pay fees; either party's liability under clause 17; a party's breach of clause 15 (Confidentiality); your breach of clause 6, clause 7 or clause 14.3; or liability for fraud, wilful misconduct, death or personal injury caused by negligence, or any liability that cannot lawfully be limited.
18.6 Allocation of risk. You acknowledge that these limits reflect the fees charged and that we would not provide the Service on these terms without them.
19.1 Neither party is liable for a failure or delay caused by an event beyond its reasonable control, including natural disaster, war, terrorism, epidemic, industrial action, failure of a public telecommunications or power network, and denial of service attacks, provided it takes reasonable steps to mitigate and notifies the other party. This does not excuse payment for services actually provided. If the event continues for more than 30 days, either party may terminate the affected services on notice.
20.1 This agreement starts when you first accept it and continues until all subscriptions have ended or it is terminated.
20.2 Termination for convenience. You may terminate at any time by cancelling all subscriptions and closing your account, effective at the end of the current paid term. We may terminate a paid plan on 60 days' notice, and will refund pre-paid unused fees on a pro rata basis. We may remove an entry from the invitation list, or close an account that has never been activated, at any time.
20.3 Termination for cause. Either party may terminate immediately if the other commits a material breach that is not remedied within 30 days of notice, or becomes insolvent, has an administrator or liquidator appointed, or ceases to carry on business.
20.4 Suspension. We may suspend an Address, a Domain, a User or the account immediately, and without the 30-day cure period, where we reasonably believe it is necessary to prevent material harm to the network, another user or us, to stop unlawful activity, or to comply with law, including for a serious AUP breach, a security compromise, or Agent behaviour that threatens the network. We will limit the suspension to what is necessary, notify you as soon as we reasonably can, and restore the Service when the cause is resolved.
20.5 On termination: your right to use the Service ends; you must pay all amounts accrued; we will make Customer Data available for export for 30 days; after that we will delete it within 90 days, except where we must retain it by law or where it exists in backups, which are overwritten on the cycle described in the Privacy Policy.
20.6 Clauses 3, 8.1, 8.4, 9.2, 14, 15, 16.3, 16.4, 17, 18, 20.5, 20.6, 22 and 23 survive termination, along with any clause that by its nature should.
21.1 We may change these Terms. For a change that materially and adversely affects you, we will give at least 30 days' notice by email to your account contact and by a notice in the Service before it takes effect.
21.2 The change applies from the effective date, or, for a month already paid for, from the start of the following month if you tell us within the notice period that you do not accept it.
21.3 If you do not accept a material change, your remedy is to cancel before it takes effect and receive a pro rata refund of any fees paid for a period after cancellation. Continuing to use the Service after the effective date is acceptance.
21.4 Changes required by law, or to the AUP to address new abuse, may take effect immediately, and we will notify you as soon as practicable.
22.1 Notices to you are given by email to your account contact, by a message to your administrative Address, or by an in-Service notice, and are taken to be received on the day sent unless a delivery failure is received.
22.2 Notices to us must be sent to [legal@agentscolab.com] and, for formal legal notices, also by post to [88 Jolimont Street, East Melbourne VIC 3002, Australia]. They are taken to be received on acknowledgement or five business days after posting.
22.3 Assignment. You may not assign or novate this agreement without our written consent, not to be unreasonably withheld, except to a successor of substantially all of your business or assets, on notice to us. We may assign on notice to a successor of substantially all of our business or assets. Any other purported assignment is void.
22.4 Subcontracting. We may use sub-processors and subcontractors and remain responsible for their performance. Sub-processors are listed at #/legal/subprocessors and governed by the DPA.
22.5 No partnership. Nothing creates a partnership, joint venture, employment or agency relationship, and neither party may bind the other.
22.6 No third-party rights. No person other than the parties may enforce this agreement, except an indemnified party under clause 17.
22.7 Entire agreement. This agreement is the entire agreement about its subject matter and replaces all earlier discussions and representations. Neither party has relied on any statement not set out here, except that nothing excludes liability for fraudulent misrepresentation. No purchase order term or vendor portal term has any effect.
22.8 Severability. If a provision is unenforceable, it is read down or severed to the minimum extent, and the rest continues.
22.9 Waiver. A failure or delay in exercising a right is not a waiver, and a single exercise does not prevent a further one.
22.10 Counterparts and electronic signature. An Order Form may be signed in counterparts and electronically.
22.11 Language. The English version of this agreement prevails over any translation.
23.1 This agreement is governed by the law of Victoria, Australia. The parties submit to the non-exclusive jurisdiction of the courts of Victoria and the courts competent to hear appeals from them, and waive any objection based on inconvenient forum.
23.2 Talk first. Before starting proceedings, a party must give the other written notice describing the dispute, and the parties must have their senior representatives try in good faith to resolve it within 30 days. This does not prevent either party seeking urgent injunctive or interlocutory relief, or recovering an undisputed debt.
23.3 If you are a consumer, nothing in this clause deprives you of the protection of the mandatory law of the country where you live, or of the right to bring proceedings there.
23.4 Each party may bring claims only on its own behalf. To the extent permitted by law, neither party may bring a claim as a plaintiff or class member in a class or representative action against the other. This clause does not affect any right that cannot be waived by law.
Channels. Email to [support@agentscolab.com] and the Address [@support@agentscolab.com]. In-Service ticketing on every plan.
Hours. 9:00 to 17:00 Australian Eastern Time, business days, excluding Victorian public holidays. Enterprise additionally receives out-of-hours response for Severity 1 incidents.
Levels. Personal and Business receive Level 2 support. Enterprise receives Level 2 and above, as stated in its Order Form. Level 1 is triage. Level 2 means your ticket is handled by an engineer who can read the delivery records for your Addresses and act on your account, rather than by a script. Level 3 is escalation to the engineers who build the Service, which Enterprise may call on directly.
Target first response, measured within support hours unless stated:
| Severity | Definition | Personal | Business | Enterprise |
|---|---|---|---|---|
| 1 | Service unavailable or a security incident, no workaround | 8 hours | 8 hours | 1 hour, 24x7 |
| 2 | Major function impaired, workaround exists | 1 business day | 1 business day | 4 hours |
| 3 | Minor fault or question | 2 business days | 2 business days | 1 business day |
| 4 | Feature request | Logged | Logged | Logged |
These are response targets, not resolution guarantees. We will use reasonable endeavours to resolve faults promptly and to keep you informed.
Out of scope. Support for your own code, your Agents' behaviour, your DNS provider, your identity provider, and third-party services.
Applies to every paid plan. Unpaid trials and features marked beta carry no service level.
Monthly Uptime Percentage = (total minutes in the calendar month minus Unavailable Minutes) divided by total minutes in the month.
Unavailable means the Service's message submission, delivery or retrieval functions return errors for all requests for a continuous period of five minutes or more, measured by our monitoring.
Commitment. 99.9% per calendar month for Personal and Business. 99.99% per calendar month for Enterprise.
Service credits, applied against the next invoice, on request made within 30 days of the end of the affected month:
| Monthly Uptime | Credit |
|---|---|
| Below the commitment but at or above 99.0% | 10% of that month's fees |
| Below 99.0% but at or above 95.0% | 25% |
| Below 95.0% | 50% |
Exclusions. Scheduled maintenance notified at least 48 hours in advance, emergency security maintenance, force majeure, faults caused by you, your Agents, your configuration, your DNS, your network or a third-party service, suspension under clause 20.4, and unavailability of a Recipient Host.
Sole remedy. Service credits are the sole and exclusive remedy for failure to meet a service level, other than any remedy that cannot be excluded by law. If Monthly Uptime is below 95.0% for two consecutive months, you may terminate the affected subscription for cause and receive a pro rata refund of any fees paid for a period after termination.
What we collect, our two roles, and your rights.
Version 1.0. Published [DATE]. Effective [DATE]. Last updated [DATE].
Details to confirm before publication. Placeholders in square brackets are for the operator to complete: legal entity name and ABN, registered address, hosting region, the sub-processor list, and the EU and UK representatives. The sub-processor page at
#/legal/subprocessorsstill has to be written and published. Everything else is drafted to stand as written, and matches what the website says.This document is a draft prepared for review. It is written to the Australian Privacy Principles, the GDPR and UK GDPR, and US state privacy law, but it has not been reviewed by a qualified privacy lawyer, and a privacy policy must describe what your systems actually do. Verify every factual statement against your build before you publish it.
Agents Colab hosts verifiable messaging addresses for people and for AI agents. Running that service means handling two very different kinds of information, and the difference decides your rights.
Account information is information about the people who buy and administer the Service. We decide how it is used. We are the controller of it.
Message content is what our customers and their agents send through the Service. Our customer decides what it is and why it is sent. We hold it and deliver it on their instructions. We are the processor of it, and if you want it changed or deleted, the customer is the right place to ask. We will help them.
Three things we do not do. We do not sell personal information. We do not use message content to train machine learning models. We do not read message content except where clause 7 says we may.
Two things to understand. Messages are delivered to other providers on the open fmsg network, and once a message reaches another provider that provider holds a copy, which our deletion of our copy does not reach. And nothing expires here: every address keeps its whole history for as long as it exists, so data stays with us until you delete it or close the account.
[Agents Colab Pty Ltd] [ABN / ACN], [88 Jolimont Street, East Melbourne VIC 3002, Australia].
| Purpose | Contact |
|---|---|
| Privacy questions, requests and complaints | [privacy@agentscolab.com], Address [@privacy@agentscolab.com] |
| Privacy Officer | [Name or role], [privacy@agentscolab.com] |
| Security reports | [security@agentscolab.com] |
| Legal notices | [legal@agentscolab.com] |
| EU representative, Article 27 GDPR | [To be appointed] |
| UK representative, Article 27 UK GDPR | [To be appointed] |
Post to the address above, marked "Attention: Privacy Officer".
This policy applies to:
It also explains, in section 6, how we handle the personal information contained in customer messages, where we act on our customer's instructions rather than our own.
We are the controller (in Australian terms, the entity that decides the purposes of collection) for: account and billing records, administrative contact details, security and audit logs, support conversations, website analytics, marketing contacts, and message metadata to the extent we use it to run, secure and bill for the Service.
We are the processor for the content of messages, attachments, files and directory entries our customers put into the Service. Our customer is the controller. Our handling of that data is governed by our contract with them and by the Data Processing Addendum at #/legal/dpa, which sets out the instructions we accept, our confidentiality obligations, our sub-processors, our security measures and our assistance with data subject requests.
If you are an individual whose personal information is in a customer's messages and you want access, correction or deletion, contact that customer first. If you cannot identify them or they do not respond, contact us and we will help you reach them, or act ourselves where the law requires us to.
| Category | Examples | Why we collect it | Legal basis (GDPR / UK GDPR) |
|---|---|---|---|
| Account | Name, work email, organisation, role, job title, username, password hash, multi-factor settings, language, time zone | Create and run your account, authenticate you, provide support | Contract, Art. 6(1)(b) |
| Domain and configuration | Domains you verify, DNS verification records, keys and key identifiers, plan and address configuration, agent address registrations and the human owner recorded for each | Deliver the Service, prove sender authenticity, enforce the Terms | Contract, Art. 6(1)(b); legitimate interests, Art. 6(1)(f) |
| Billing | Billing contact, billing address, tax identifiers, plan, invoices, payment status, last four digits and card brand from our payment processor | Charge you, keep tax records, prevent payment fraud | Contract, Art. 6(1)(b); legal obligation, Art. 6(1)(c) |
| Message metadata | Sender and recipient addresses, timestamps, message and thread identifiers, integrity hashes, size, delivery status and errors, host of the counterparty | Route and deliver messages, retry, detect abuse, meter usage, maintain the integrity chain, investigate incidents | Contract, Art. 6(1)(b); legitimate interests in network security and abuse prevention, Art. 6(1)(f) |
| Usage and technical | IP address, user agent, device and browser characteristics, pages and API endpoints called, request volumes and rates, error traces, session and login records | Operate, secure and improve the Service, diagnose faults, enforce limits, detect intrusion | Legitimate interests, Art. 6(1)(f) |
| Support | Your messages to us, tickets, call notes, screenshots and logs you send, satisfaction responses | Answer you and improve the Service | Contract, Art. 6(1)(b); legitimate interests, Art. 6(1)(f) |
| Marketing | Work contact details, subscription and preference records, whether an email was opened or a link clicked, event attendance | Send you material you asked for, or relevant material about the Service to existing business contacts | Consent, Art. 6(1)(a), where required; legitimate interests, Art. 6(1)(f), otherwise |
| Website | Cookie and similar identifiers, referrer, aggregate analytics | Run the site, remember your choices, understand usage | Consent for non-essential cookies; legitimate interests for essential ones |
| Invitation list | Your name, work email, company, the domain you want to use, the first address you want, how many people and agents you have, which agents you already run, what you want them to talk about, the page you came from, the country our network provider reports, a salted hash of your IP address, and your browser's user agent string | Run the invitation list in the order people joined it, decide who to invite next, hold the address you asked for until your invitation comes, and stop automated submissions | Consent, Art. 6(1)(a); legitimate interests in running the beta and preventing abuse, Art. 6(1)(f) |
| Compliance and safety | Abuse reports, investigation records, sanctions and fraud screening results, records of legal requests | Meet legal obligations, protect the network, establish and defend legal claims | Legal obligation, Art. 6(1)(c); legitimate interests, Art. 6(1)(f) |
Where we rely on legitimate interests, we have weighed our interest against your rights and concluded it does not override them. You may object at any time under section 11, and we will stop unless we have compelling grounds or need the data for legal claims. We will give you our balancing assessment on request.
The invitation list. Only your email address is required. We do not pass what you put on that list to anyone else, we do not use it to market anything other than the invitation, and we do not take payment details there. Ask us at [privacy@agentscolab.com] and we will remove your entry and delete it.
Sensitive information. We do not seek special category or sensitive information for the purposes above. Do not send it to us in support tickets. If it reaches us in message content, we handle it as our customer's data under section 6.
If you do not provide it. Account, domain and billing information is necessary to provide the Service. Without it we cannot open or run an account. Other information is optional.
6.1 We store message bodies, attachments and files so we can deliver them, hold them for as long as the address exists, and let the customer search, export and audit them. Nothing expires. Content stays until the customer deletes it or the account closes, on every plan.
6.2 We do not read message content, and no member of our staff accesses it, except in the circumstances in section 7.
6.3 We do not use message content to train, fine-tune or evaluate machine learning models, ours or anyone else's, and we do not disclose it for anyone's advertising or marketing.
6.4 Addresses are meant to be found. An address in the form @name@domain is designed to be verifiable and reachable by other participants on the network. Where a customer opts into the public directory, the entries they choose are published. Do not put anything in an address or a directory entry that you do not want visible.
6.5 Delivery discloses data by design. Sending a message to an address on a domain we do not host means transmitting it, and the personal information in it, to the recipient's provider. We cannot control what that provider or that recipient then does. Our customer decides who to send to.
6.6 Deletion has a boundary. When a customer deletes a message, we delete our copy on the schedule in section 10. Copies already delivered to other providers are outside our control and are not deleted by that action.
6.7 Encryption. Content is encrypted in transit between hosts and at rest in our storage. We hold the keys to storage encryption, which means we are technically capable of accessing content, which is why section 7 exists and is narrow.
6.8 Automated threat screening. On the Enterprise plan, every message is screened automatically for malware and other threats before it is delivered. That screening is done by software, on delivery, and produces a result about the message, not a copy of it for anyone to read. A person looks at a message only where section 7 allows it.
6.9 Closed by default. An address exchanges messages only inside its own domain until the customer allows another domain or another address. Every incoming message is checked against that setting before it is stored.
This section is about a person reading message content. Automated screening under section 6.8, delivery, storage, indexing for the customer's own search, and abuse detection carried out by software are not covered by it. A member of our staff accesses message content only where one of these applies, only to the minimum extent necessary, only where authorised, and always with an access log the customer may request:
(a) the customer instructs or authorises us to, for example to investigate a support ticket; (b) it is strictly necessary to diagnose or fix a fault in delivery or storage that we cannot resolve from metadata; (c) it is necessary to detect, prevent or stop a security incident, malware, fraud or an attack on the network; (d) it is necessary to investigate a credible report of a serious breach of the Acceptable Use Policy; (e) we are legally required to, and section 9 applies; or (f) there is an imminent risk of death or serious harm to a person.
8.1 The website. agentscolab.com sets no cookies, runs no analytics and carries no third-party trackers. The invitation form sends us what you type into it and nothing else.
8.2 The Service. Signing in sets strictly necessary cookies for your session, authentication, load balancing and security. These cannot be switched off without breaking the Service, and no consent is required for them. Your theme, language and dismissed notices may be stored in your own browser.
8.3 We do not use advertising cookies, we do not permit third-party advertising trackers, and we do not build cross-site profiles. If we add analytics later we will describe it here, ask for consent where consent is required, and publish the notice before we turn it on.
8.4 You can block cookies in your browser, though the Service may not work. We honour Global Privacy Control signals where they are received.
9.1 Sub-processors and service providers. Cloud hosting and storage, email delivery, payment processing, error monitoring, customer support tooling and security services. Each is bound by contract to protect the data, to use it only for us, and to meet the standards in the DPA. Cloudflare hosts the website and stores the invitation list. The current full list, with entity, purpose and location, is at #/legal/subprocessors. Customers may subscribe to change notices there and object to a new sub-processor under the DPA.
9.2 Other providers on the network. As described in section 6.5.
9.3 Your own organisation. If you use an account provided by an organisation, its administrators can see and manage your account, your addresses, your usage and the messages in that account.
9.4 Professional advisers, under duties of confidence.
9.5 Corporate transactions. In a merger, acquisition, financing or sale of assets, subject to confidentiality, and continuing under this policy or one materially as protective. We will notify you of a change of controller.
9.6 Law enforcement, regulators and courts. We disclose only where we are legally required to, or where section 7(f) applies. Our practice is to require valid legal process, to interpret requests narrowly, to refuse or challenge those that are overbroad or unlawful, and, unless legally prohibited or there is a risk to life, to notify the affected customer before disclosing so they can respond. We publish periodic transparency reporting on the volume and type of requests received.
9.7 We do not sell personal information, and we do not share it for cross-context behavioural advertising, as those terms are used in United States state privacy laws. We do not pass the details you put on the invitation list to anyone else.
| Data | Retention |
|---|---|
| Message content and attachments | Kept for the life of the address, on every plan. Nothing expires. Deleted when the customer deletes it, or after the account closes, as clause 20.5 of the Terms provides |
| Message metadata and integrity records | Kept with the messages they belong to, then up to 12 months after the account closes, in reduced form, for security, abuse and billing dispute purposes |
| Invitation list entries | Until the account is opened, or until you ask us to remove you, and in any case no more than 24 months after your last contact with us |
| Account records | For the life of the account, then 12 months after closure |
| Billing, invoicing and tax records | 7 years, as required by Australian tax law |
| Security and access logs | 12 months, longer where an investigation is open |
| Support tickets | 3 years from resolution |
| Marketing records | Until you unsubscribe, then a suppression record kept indefinitely so we do not contact you again |
| Backups | Encrypted, rolling, overwritten within [35] days |
| Data under legal hold | For as long as the hold applies |
Deleted data is removed from live systems promptly, and from backups as those backups are overwritten. We may keep de-identified or aggregated data indefinitely, and we do not attempt to re-identify it.
11.1 Subject to the law that applies to you, you may:
11.2 How to exercise them. Write to [privacy@agentscolab.com] or use the controls in your account. We may need to verify your identity, and will ask for no more than necessary to do so.
11.3 How long we take. Within 30 days for Australian Privacy Principle requests, and within one month for GDPR and UK GDPR requests, extendable by two further months for complex requests with notice to you and reasons. We do not charge, unless a request is manifestly unfounded or excessive, in which case we will tell you the reasonable cost before proceeding, or refuse and explain.
11.4 If we refuse, we will tell you in writing, give reasons and explain how to complain.
11.5 US state privacy rights. Residents of California and other US states with comprehensive privacy laws have rights to know, delete, correct and, where applicable, opt out of sale, sharing and targeted advertising and certain profiling. We do not sell or share personal information, so there is nothing to opt out of. We will not discriminate against you for exercising a right. You may use an authorised agent, with proof of authority.
12.1 Tell us first, at [privacy@agentscolab.com]. We will acknowledge within 5 business days and respond substantively within 30 days.
12.2 If you are not satisfied, you may complain to a regulator:
You may complain to a regulator without coming to us first, though we would rather have the chance to fix it.
13.1 We host the Service in [region to be confirmed]. Some sub-processors, and our own staff, are located elsewhere, including [countries to be confirmed on completion of the sub-processor list].
13.2 Where we transfer personal data out of the EEA, the UK or Switzerland, we rely on an adequacy decision where one covers the destination, and otherwise on the European Commission's Standard Contractual Clauses, with the UK International Data Transfer Addendum where the UK GDPR applies, together with a transfer impact assessment and supplementary technical measures including encryption in transit and at rest.
13.3 Under Australian Privacy Principle 8, we take reasonable steps to ensure that an overseas recipient does not breach the Australian Privacy Principles, principally through binding contractual terms in the DPA.
13.4 You may request a copy of the transfer mechanism relied on, with commercial terms redacted.
14.1 Our measures include encryption in transit and at rest, least-privilege and role-based access control, multi-factor authentication for staff, segregated environments, centralised logging and monitoring, vulnerability management and patching, secure development practices with code review, background checks and confidentiality obligations for staff, vendor security review, and a tested incident response plan. The current technical and organisational measures are in Annex II to the DPA.
14.2 No system is perfectly secure. Transmitting data to us is at your own risk to the extent the law permits, and section 16.3 of the Terms applies.
14.3 Breach notification. Where a data breach is likely to result in serious harm, we notify the Office of the Australian Information Commissioner and affected individuals as required by the Notifiable Data Breaches scheme, and we notify affected supervisory authorities within 72 hours and affected individuals without undue delay where the GDPR or UK GDPR applies. Where we act as processor, we notify the customer without undue delay after becoming aware, so they can meet their own obligations.
14.4 Report a vulnerability to [security@agentscolab.com]. We will not pursue legal action against good-faith security research that is reported to us there, does not degrade the Service for anyone else, and does not access or retain another customer's data.
The Service is for business use and is not directed at children. We do not knowingly collect personal information from anyone under 16. If we learn that we have, we will delete it. Contact [privacy@agentscolab.com] if you believe a child has provided information to us.
16.1 We send commercial electronic messages only with consent, express or inferred, as the Spam Act 2003 (Cth) allows, and where the GDPR or UK GDPR applies, only on the basis stated in section 5.
16.2 Every marketing message identifies us and carries a working unsubscribe. We action unsubscribes within 5 business days and usually immediately.
16.3 Service messages about your account, billing, security and material changes are not marketing, and you cannot unsubscribe from them while you hold an account.
We may update this policy. For a material change, we will give at least 30 days' notice by email to account contacts and by a notice in the Service before it takes effect, and we will keep prior versions, which you may ask us for at [privacy@agentscolab.com]. The version and date at the top of this page always show the current one. Continuing to use the Service after a change takes effect means you accept it, except where a change requires consent, in which case we will ask.
Personal information and personal data are used interchangeably and mean information about an identified individual, or an individual who is reasonably identifiable.
Controller and processor have their GDPR meanings and are used to describe an equivalent division of responsibility under Australian law, which does not use those terms.
Where a law that applies to you gives you a right that this policy does not describe, you have that right, and this policy does not limit it.
What may not be sent, and what happens if it is.
Version 1.0. Effective [DATE].
This policy forms part of the Terms of Service. It applies to every user, every address and every agent on the Service. Where it is unclear whether something is allowed, ask us at [support@agentscolab.com] before doing it.
The Service exists so that a message from an address can be trusted to have come from that address, and so that a record of what was said can be relied on afterwards. Anything that erodes that trust is prohibited, whether or not it appears below.
You must not use the Service to send, store, request or facilitate:
Deception and impersonation - Impersonating a person, organisation, domain or address, or misrepresenting where a message came from. - Presenting an AI agent as a human being where that would mislead the recipient or breach a law requiring disclosure of automated interaction. - Phishing, credential harvesting, business email compromise, invoice fraud, or any attempt to induce a payment or disclosure by deception. - Forging headers, tampering with integrity chains, or attempting to make a record appear other than as it is.
Unsolicited messaging - Sending commercial electronic messages without the consent required by the Spam Act 2003 (Cth), the CAN-SPAM Act, the ePrivacy rules or the equivalent law where your recipient is. - Bulk or automated messaging to addresses that have not opted in, harvested addresses, or purchased lists. - Sending without a functioning unsubscribe where one is required, or after an unsubscribe. - Using multiple addresses, agents or accounts to evade a block, a rate limit or a filter.
Malicious activity - Malware, ransomware, spyware, exploits, or links or attachments intended to deliver them. - Scanning, probing, penetration testing (without our prior written consent), denial of service, or any attempt to gain unauthorised access to any system, account, address or data. - Circumventing authentication, rate limits, quotas, retention settings or billing. - Using the Service as infrastructure for command and control, botnets, cryptomining, proxying or traffic laundering.
Unlawful and harmful material - Anything unlawful where you are, where we are, or where your recipient is. - Child sexual abuse material. We report this to the relevant authorities without notice to the account holder. - Content that incites violence, promotes terrorism, or facilitates trafficking in people, weapons or controlled substances. - Harassment, threats, stalking, doxxing, or the non-consensual distribution of intimate images. - Content that infringes copyright, trade marks, patents, trade secrets or moral rights. - Personal information handled in breach of privacy or data protection law, including sensitive information you have no lawful basis to send.
Prohibited uses - Life-critical uses as described in clause 6.6 of the Terms. - Reselling the Service, or providing it to a third party as your own service, without our written agreement. - Benchmarking for publication without our consent, or use to build a competing product. - Evading sanctions or export controls, or providing the Service to a sanctioned person or jurisdiction.
Plans include allowances rather than hard technical ceilings. If an address goes past its daily message allowance or its 32 GB of storage, we get in touch with you. We do not stop delivery and we do not drop messages for that reason alone. Use that is materially disproportionate to your plan, degrades service for others, or is designed to shift cost onto shared infrastructure, may be rate limited after notice, or immediately where it threatens the Service. This is separate from the limits you set yourself on an address, which stop that address as you intended.
Report abuse to [abuse@agentscolab.com] or the address [@abuse@agentscolab.com], with the message identifiers and, where you can, the full message including headers. We acknowledge reports and act on those we can verify. We do not usually tell the reporter what action we took.
Depending on severity we may warn you, rate limit, suspend an address, an agent, a domain or the account, remove content where we are required or permitted to, terminate under clause 20.3 of the Terms, or report to authorities. Suspension for serious breach is immediate under clause 20.4. We aim to act proportionately, to tell you what happened and why, and to restore service when the cause is resolved. Repeated or deliberate breach ends the account.
We may update this policy to address new forms of abuse, effective immediately, with notice as soon as practicable, as clause 21.4 of the Terms provides.
Processor terms, standard contractual clauses, security measures.
Version 1.0. Effective [DATE].
This DPA forms part of the Terms of Service between [Agents Colab Pty Ltd] (Processor, we) and the Customer (Controller, you). It applies whenever we process personal data on your behalf. If you need a signed copy, email [legal@agentscolab.com].
Draft for legal review. The Standard Contractual Clauses must be attached in full with the module elections and annexes completed before this is offered to EEA or UK customers.
1.1 You are the controller (or processor for your own customer) of personal data contained in Customer Data. We are the processor (or sub-processor).
1.2 We are an independent controller of Service Data, account, billing and security data, as described in our Privacy Policy, and we process that data for our own legitimate purposes of running, securing and billing for the Service.
1.3 You are responsible for the lawfulness of the data you submit, for the notices and legal bases required for it, and for the instructions you give us.
2.1 We process Customer Data only on your documented instructions, which are: these Terms, this DPA, the configuration you set in the Service, and any further written instruction you give that is consistent with the Service.
2.2 We will tell you if, in our opinion, an instruction breaches applicable data protection law, and may suspend performance of that instruction. We may process where required by law that applies to us, and where permitted will tell you first.
Our personnel who access Customer Data are bound by confidentiality obligations that survive their engagement, receive privacy and security training, and are granted access on a least-privilege, need-to-know basis, logged and reviewed.
We implement and maintain the technical and organisational measures in Annex II, appropriate to the risk, and will not materially reduce them during your subscription term.
5.1 You give general authorisation for us to engage sub-processors. The current list is at #/legal/subprocessors.
5.2 We will give at least 30 days' notice before adding or replacing a sub-processor, by email to subscribers of that page. You may object on reasonable data protection grounds within that period. We will work with you to find a resolution, and if we cannot, you may terminate the affected subscription and receive a pro rata refund of pre-paid unused fees, as your sole remedy.
5.3 We impose data protection obligations on each sub-processor that are no less protective than this DPA, and we remain fully liable to you for their performance.
6.1 The Service gives you tools to access, correct, export and delete Customer Data yourself.
6.2 Where a data subject contacts us directly about Customer Data, we will not respond substantively, and will refer them to you and tell you promptly, unless legally required to respond.
6.3 We will give you reasonable assistance with requests you cannot fulfil through the Service, at no charge for reasonable volumes.
7.1 We will assist you, taking into account the nature of processing and the information available to us, with your obligations on security, breach notification, data protection impact assessments and prior consultation.
7.2 We will make available the information reasonably necessary to demonstrate compliance with this DPA, in the first instance through our security documentation, questionnaire responses and any third-party audit report or certification we hold.
7.3 Where that is not sufficient for your regulator or applicable law, you may audit us, or appoint an independent auditor who is not our competitor and who signs confidentiality obligations, once in any 12 months (and any time following a confirmed breach affecting your data), on 30 days' notice, during business hours, without unreasonable disruption, and at your cost.
We will notify you without undue delay, and in any case within 48 hours, after becoming aware of a personal data breach affecting Customer Data. The notice will describe the nature of the breach, the categories and approximate numbers affected where known, the likely consequences, the measures taken or proposed, and a contact point. We will provide updates as we learn more, and will not delay initial notification for the sake of completeness.
9.1 We process Customer Data in [region to be confirmed] and transfer it only as described in Annex III.
9.2 For transfers from the EEA, the Standard Contractual Clauses (Commission Implementing Decision (EU) 2021/914) are incorporated, with Module Two (controller to processor) or Module Three (processor to processor) as applicable, docking clause included, Clause 9 Option 2 with 30 days' notice, Clause 11 optional redress body not selected, Clause 17 governed by Irish law, Clause 18(b) courts of Ireland. The Annexes to this DPA populate Annexes I, II and III of the SCCs.
9.3 For transfers from the UK, the ICO International Data Transfer Addendum applies, with Tables 1 to 4 completed by reference to this DPA and Table 4 election "neither party".
9.4 For transfers from Switzerland, the SCCs apply with references to the GDPR read as references to the FADP, the Federal Data Protection and Information Commissioner as competent authority, and "member state" not limiting the rights of data subjects in Switzerland.
9.5 Australian Privacy Principle 8 is addressed by the contractual obligations in this DPA.
On termination we make Customer Data available for export for 30 days, then delete it within 90 days, and from backups as those backups are overwritten within [35] days, except where retention is required by law. We will certify deletion on written request.
If we receive a legally binding request from a public authority for Customer Data, we will, unless legally prohibited, notify you promptly and suspend the disclosure while we seek a waiver or challenge the request. We will disclose only the minimum required, we will challenge requests that appear unlawful or overbroad, and we keep records of requests received and our responses.
Liability under this DPA is subject to the limitations in clause 18 of the Terms, except where applicable data protection law prevents that. If this DPA conflicts with the Terms on the processing of personal data, this DPA prevails. If it conflicts with the SCCs, the SCCs prevail.
Parties. Data exporter: the Customer, contact details as in the account. Data importer: [Agents Colab Pty Ltd], [address], [privacy@agentscolab.com].
Subject matter. Provision of the hosted fmsg messaging service.
Duration. The subscription term, plus the deletion periods in clause 10.
Nature and purpose. Hosting, transmission, delivery, storage, retention, indexing, search, export, backup, security monitoring and support.
Categories of data subject. The Customer's personnel and contractors, the accountable owners of agents, the Customer's own customers, suppliers and counterparties, and any other individual whose personal data the Customer includes in a message.
Categories of personal data. Identity and contact details, addresses, employment and role information, message content and attachments chosen by the Customer, metadata, and any other data the Customer submits. An address may be held by a person or by an agent, and both are charged and treated the same way.
Special category data. Not requested and not expected. Where the Customer submits it, the additional restrictions are: encryption at rest and in transit, access restricted to named personnel, and access logged and available to the Customer.
Frequency. Continuous.
Retention. Message content and attachments are kept for the life of the address, on every plan, and do not expire. Deletion on termination is as set out in clause 10 of this DPA and section 10 of the Privacy Policy.
Sub-processors. As listed at #/legal/subprocessors, for the purposes and durations stated there.
Competent supervisory authority (SCC Annex I.C). [To be completed by the Customer, or the authority of the Customer's member state of establishment, or of the representative, or of the data subjects, as Clause 13 provides.]
#/legal/subprocessors.The current sub-processor list, with entity name, purpose, processing location and transfer mechanism, is published at #/legal/subprocessors and is incorporated into this Annex. It includes Cloudflare, which hosts the website and stores the invitation list. [Populate before publication.]