StandIn SDK

About  ·  How It Works  ·  Quickstart  ·  Plugins  ·  Agent Skills  ·  Documentation

About

StandIn is the open-source SDK that gives your AI agent a seat in Microsoft Teams. It answers calls,
joins meetings, sees the shared screen, and replies in chat, as a real participant in your own tenant.

You build the agent. StandIn handles everything between it and Microsoft Teams: joining the call, the Microsoft
side, the media, and the avatar tile. There is no Microsoft Teams SDK to learn, no Graph API, and no media
infrastructure to run.

• Build with one API in Python or TypeScript. Same surface, same contract, at parity.
• Connect what you already run: ElevenLabs, Deepgram, Cartesia, OpenAI, LiveKit, Hermes
  Agent, OpenClaw, or your own agent in about 80 lines.
• Try it before writing any agent code. The built-in echo answers a real call and sends your own
  voice back.

How It Works

Three parts, and you only write the last one.

   Microsoft Teams  <-->  StandIn  <-->  your worker  <-->  your agent

The SDK is the transport. Speech recognition, speech generation and reasoning stay with your framework
or provider, so you keep the brain and StandIn never sees your model keys.

What it does so you do not have to: authenticate the connection, keep one live call per caller, pace the
audio in both directions, stop the agent talking over someone who interrupts, and shut down cleanly when
a call ends or a provider fails.

It also gives your agent the things that belong to being on a call rather than to any provider, so each
plugin gets them without writing them again:

The documentation covers the wire protocol and audio format when you need them.

Quickstart

The built-in echo plugin answers a real Microsoft Teams call and sends your voice back, so you can prove
the connection before adding an agent. It needs no framework and no API key.

python3 -m venv .venv
source .venv/bin/activate
pip install standin-sdk

export STANDIN_SECRET="your-StandIn-connection-secret"
python -m standin.plugins.echo

Or in TypeScript, the same echo, the same wire protocol:

npm install @komaa/standin-sdk

export STANDIN_SECRET="your-StandIn-connection-secret"
npx standin-echo

Create an identity at standin.komaa.com, expose port 9442 through a tunnel
or ingress, and register the public wss:// URL with the /msteams/calling path as that identity's
agent voice URL. Then call it from Microsoft Teams and hear yourself.

Once that works, swap the echo for your agent. The whole contract is seven callbacks, and you implement only
the ones you need.

Python

from standin import CallServer, CallSession


class MyAgent:
    async def on_start(self, session: CallSession) -> None:
        self._call = session

    async def on_caller_audio(self, pcm: bytes) -> None:
        await self._call.send_audio(await my_framework.respond(pcm))   # your agent


server = CallServer(handler_factory=MyAgent)
await server.start()

TypeScript

import { CallServer, type CallSession } from "@komaa/standin-sdk";

class MyAgent {
  #call!: CallSession;

  async onStart(session: CallSession) {
    this.#call = session;
  }

  async onCallerAudio(pcm: Buffer) {
    await this.#call.sendAudio(await myFramework.respond(pcm)); // your agent
  }
}

await new CallServer({ handlerFactory: () => new MyAgent() }).start();

Plugins

Every plugin ships inside the one package, so a new capability lands once and all of them get
it. Each one lives in libraries/python/standin/plugins/ or
libraries/typescript/src/plugins/, and has a runnable example under examples/.

Most of them cost nothing to install. ElevenLabs, Deepgram and Cartesia are reached over an
ordinary WebSocket in both languages, and OpenAI Realtime the same way in TypeScript, so they
are already there after pip install standin-sdk or npm install @komaa/standin-sdk, with no
extra dependency at all.

Only a plugin that runs a framework inside your process asks for more, and it says so in
one line: pip install "standin-sdk[livekit]" or pip install "standin-sdk[hermes-agent]" in
Python, or the optional peer packages named by the plugin in TypeScript.

Missing yours? Adding one is a copy of the echo plugin plus a few lines of glue. See
CONTRIBUTING.md.

Agent Skills

A coding agent can install these and wire StandIn for you: connection secret, OpenClaw,
Hermes Agent, and publishing /msteams/calling.

npx skills add komaa-com/skills

Install one skill with npx skills add komaa-com/skills --skill standin-openclaw.
The skills live in komaa-com/skills, not in this repository.

Documentation

Full guides, the call handler reference, the audio and chat lanes, and the security model live at
docs.komaa.com.

Demo

A real Microsoft Teams call: the agent answers, sees the shared screen, speaks when addressed, and appears as a
lip-synced avatar.

Contributing

Issues and pull requests are welcome, and a new plugin is the most useful thing you can add.
make check needs no API keys, no Microsoft tenant and no StandIn account, so a pull request from a fork
passes CI without any secrets. See CONTRIBUTING.md.

License

MIT, copyright 2026 Komaa DigiTech.

StandIn is independent software. It is not affiliated with, endorsed by, or sponsored by Microsoft.
Microsoft and Microsoft Teams are trademarks of the Microsoft group of companies.