The API Channel Is Now Live: Message Your AI Employees From Your Own Software

Your own software can now message an AgentTeams employee directly and get a real reply back. Point your product, an internal tool, or a script one of your developers wrote at a specific employee, and it answers using its full set of instructions, company knowledge, connected tools, and memory of the conversation, the exact same employee that already answers your team in Slack or the dashboard, not a stripped-down version built for outside callers.

We made the case for this months ago in AI agents beyond Slack: an employee is only as useful as the places it can be reached, and email, tickets, and chat each earned their own channel one at a time. This is the same idea, pointed inward at whatever your own company already runs. This article covers what actually shipped, how to turn it on, and where it fits next to everything else already in the catalog.

What is live

Here is what actually shipped:

• A direct line to any employee. An admin creates a key from Settings → API, and that one key can reach any employee in your company. You choose which one with every message you send, so there is no separate key to manage per employee.
• Full memory, per conversation. Give a message a conversation id, an email thread, an order number, a chat session, anything you already use to tell threads apart, and the employee remembers everything said in that thread the next time you write back, exactly like a Slack channel keeps its own history.
• It reads what you send it. Attach a screenshot and the employee looks at it. Attach a PDF, a spreadsheet, or a Word document and it reads it. Up to five images and ten files per message, and audio gets transcribed first.
• An answer now, or the moment it is ready. Hold the line open for up to ten minutes if your flow needs the reply immediately, or give us a web address in your own system and we deliver the finished reply there ourselves as soon as it is done, signed so your system can confirm it really came from us. Either way, you can also just check back and ask whether it is done yet.
• New kinds of requests show up on their own. Nothing to register in advance. The first time your software sends a new kind of request, say a support email versus a new order, it just works, and it quietly appears on Settings → API with a "New" tag. From there an admin can promote it onto the employee's task board to add standing instructions, or switch it off, entirely optional and never something that blocks the request itself.

Getting a key

Setup takes about two minutes. Only an admin on your team can create or revoke a key, and there is no vendor to sign in to, since you are connecting your own software to your own employees:

1. An admin opens Settings → API and clicks Create key. No developer portal, no approval queue to wait on.
2. The key and a signing secret used to verify replies appear exactly once. Copy both somewhere safe, the same as a password. If a key ever leaks, or someone leaves, revoking it from that same page takes effect immediately.
3. Copy the ID of the employee you want to reach, listed right there on the same page along with working examples to copy, and send your first message.

That is the entire setup. The same page that creates the key also lists every employee in your company and hands you a starting example, so there is nothing else to install and nobody else to wait on.

Real use cases, copy these

The channel earns its keep on the requests that used to need a person in the middle relaying them:

• Answer your own support inbox instantly. If your ticketing tool is not one we connect to directly, point its inbound handler at the API channel instead, use the thread id as the conversation id, and every reply comes back in context, the same result our Help Scout support agents already deliver, just for a tool we do not have a built-in connection to yet.
• Put an employee behind your own product's chat box. The box does not need its own AI logic bolted on. It sends the customer's message to an employee and shows back whatever comes back, so the answer draws on the same company knowledge and rules as everything else that employee already handles.
• Turn a script into a coworker. A nightly job that emails you a report, or a script watching for something odd in your data, can hand what it finds to an employee instead of a person, and get back a written response to act on or forward.
• Keep every caller separate without extra setup. One key can serve a hundred different customers or a hundred different order threads at the same time. As long as each one gets its own conversation id, the employee never mixes up who said what.

Common questions

• Do we need a developer to set this up? Someone comfortable writing a little code, yes. If the tool on the other end is something we already connect to directly, use that instead; this channel is for when the software on the other end is yours.
• Will it act completely on its own? No differently than anywhere else in AgentTeams. Whatever normally needs a human's sign-off still needs it here. If the employee is waiting on approval, your software is told that plainly instead of getting nothing back.
• Is a leaked key dangerous? Revoking one takes effect immediately, and each key is capped at 60 messages a minute, so even a mistake in your own code cannot flood an employee. And if whatever is calling us retries the same event, which email systems do constantly, tag it with an id of your own and the employee answers it once, not twice.
• Do we have to tell you in advance what our software will send? No. The first message of a new kind just works. It shows up on Settings → API afterward so an admin can see it, add standing instructions, or turn it off. Review is a courtesy, not a gate.
• Is this the same as an employee reacting to something happening in a connected tool? No. Reactions fire when something happens inside a tool you have connected, a new ticket, a new order, with no answer expected back in your system. This channel is the other direction: your own software asks a direct question and gets a direct answer.

Where this fits

The API channel is the other half of a story we started with the MCP connector. That one lets an employee reach out to almost any tool you connect it to. This one lets almost any piece of software reach in and talk to an employee. Between the two, the tools already in the catalog stop being the only door in or out.

None of this loosens anything else AgentTeams already promises. Keys are scoped to your organization, every reply we send back to your software is signed, and access can be cut off in seconds, the same posture we take with agent security everywhere else in the product.

Ready to try it? Open Settings → API in your dashboard, click Create key, and send your first message with the working example already sitting on that page. New to AgentTeams? Start with the first-agent onboarding guide.