Your AI teammate in Microsoft Teams calls.
Every meeting tool built in the last two years does the same thing: it sits outside the room and summarizes it afterward. The notetaker joins as a silent bot, records, and emails you a recap. Copilot reads the transcript once the call is over. Useful, but always one step removed from the moment that actually mattered, the moment someone asked a question and needed an answer right then.
We wanted the opposite. We wanted the agent in the room.
Today we are launching StandIn, a service that brings your AI agent into Microsoft Teams as a real participant. It joins the call under your own Microsoft identity, listens in real time, watches the shared screen, speaks when it is spoken to, and gets work done while the meeting is still happening. Not a transcript. A teammate.
The problem with "meeting AI"
Think about what a genuinely useful colleague does in a call. They follow the conversation. They notice when the slide on screen contradicts what someone just said. They stay quiet until you turn to them and ask, and then they give you a straight answer. Afterward they go and do the thing you asked, and they come back with a result.
Notetakers do none of that. They are a tape recorder with a summarizer bolted on. The intelligence, if there is any, arrives after everyone has already moved on. For a status meeting that is a mild inconvenience. For a decision meeting it is a missed opportunity, because the one moment where an agent could have pulled a number, checked a system, or answered a question is exactly the moment it was not allowed to participate.
StandIn closes that gap by treating the agent as a first-class attendee.
What StandIn actually does
It listens and talks back. Natural speech to speech, in real time, inside the roughly 200 millisecond window people expect from a live conversation. When you interrupt it, it stops, the same way a person would.
It sees the meeting. Continuous vision over screen share and camera, with per-participant attribution, so a question like "what do you make of this?" simply works. The agent knows what "this" is because it is looking at the same screen you are.
It speaks only when addressed. In a group call, StandIn stays silent until it is called by name. Then it steps in, answers, and steps back. No talking over people, no awkward interjections. It can also take a longer task away in the background and call back with the result once it is done.
It is governed by default. Recording-consent gating, a closed-by-default allowlist, and signed, encrypted media. It was built for regulated and sovereign environments from the first commit, not retrofitted for compliance later.
And it runs under your Microsoft identity. StandIn is independent software. It is not a Microsoft product and is not affiliated with or endorsed by Microsoft. We provide the bridge that lets your agent join the call, and no meeting recordings are ever stored with us.
For developers: integrate tonight
Here is the part we are most proud of. Connecting your agent is not a project. It is an evening.
StandIn speaks to your agent through a small, open bridge. The bridge handles the hard, unglamorous parts of live media, the HMAC-authenticated handshake, replay protection, back-pressure so a slow agent never floods the pipe, and clean call teardown when the meeting ends. Your agent just answers a call.
The shape of it is roughly this. You point the bridge at your agent, you point StandIn at the bridge, and StandIn does the Teams side:
Every bridge is hardened the same way, so whichever backend you choose, you inherit the same authentication, replay protection, bounded back-pressure, and graceful teardown. All bridges are open source at github.com/komaa-com/standin, and the deep documentation, with copyable quickstarts and diagrams for all seven, lives at standin.komaa.com.
Avatar, not just audio
An agent that only talks is a voice on a speakerphone. StandIn can also relay your agent's avatar as live video into the call, so when it speaks there is a face on the tile, not an empty square. For agent platforms that render a talking-head avatar, that video streams straight into Teams alongside the audio.
Sovereign by design
A lot of "AI in your meetings" quietly means "your meetings in someone else's AI." That was a non-starter for the organizations we built this for.
StandIn stores no meeting recordings. The agent joins under your own Microsoft identity inside your own tenant. Media is signed and encrypted end to end across the bridge, access is closed by default and allowlisted, and recording consent is gated rather than assumed. If you run in a regulated or sovereign environment, StandIn was designed to live there, not to ask you for an exception.
What this changes
The teams testing StandIn are not using it to summarize meetings. They are using it to shrink them. A status meeting that used to take forty five minutes and ten people becomes a five minute check-in with the two who actually need to be there, because the agent already gathered the updates, already has the numbers on screen, and already answered the routine questions before anyone dialed in.
That is the real shift. Not better notes after the meeting. Fewer meetings, and sharper ones, because the intelligence is in the room while decisions are being made.
Try it
There are three ways to start, from zero commitment to production:
- Sandbox. A shared demo bot you can call from Teams with no Azure setup at all, just to feel what a live agent in a call is like.
- Free tier. Bring your own bot identity and your own agent, and get a real, capped daily slot to develop against. This is the path for developers who want to build and test their own integration end to end.
- Paid. Bring your own Microsoft identity and your OpenClaw, Hermes, or any of the supported agents. We bring the bridge, plus an automated test call at the end of onboarding. Hours of guided setup, not weeks.
Top comments (0)