β οΈ IMPORTANT: This is an experimental project managed by a v-team from the Artificial Intelligence Business Applications Specialist Team (AIBAST), not an officially supported Microsoft product.
π Browse the Agent Library at microsoft.github.io/aibast-agents-library Β· Install the Brainstem Β· Why a local learning and prototyping tool?
Industry agent templates, the RAPP production methodology, and the RAPP Brainstem: a local frontier learning and rapid prototyping tool powered by GitHub Copilot. The Brainstem is where you learn the agent pattern and prove a prototype, not a product you ship; every skill built here is meant to graduate into GitHub Copilot, Copilot Studio, and Microsoft 365 Copilot. No provider API key or cloud setup is required for core chat beyond a GitHub account with Copilot access.
Production Guide | Microsoft AI Academy | Agent Library | Library Metrics | Brainstem API and configuration
Recommended: VS Code with GitHub Copilot. Not required, but it's the best way to explore the AI frontier: Copilot installs the Brainstem, opens it in VS Code's browser and guides you through it. Paste the prompt in docs/vscode-install-prompt.md into GitHub Copilot Chat in VS Code (agent mode), or copy it from the installer page. Don't have VS Code? Install it and set up GitHub Copilot.
Advanced: run the one-liner yourself. A fully supported alternative that needs no VS Code.
curl -fsSL https://github.com/microsoft/aibast-agents-library/releases/download/installers/install.sh | bash
Windows (PowerShell β works on factory Windows 11):
irm https://github.com/microsoft/aibast-agents-library/releases/download/installers/install.ps1 | iexAuto-installs Python 3.11, Git, and GitHub CLI via winget if missing.
Then:
brainstem # start the server β localhost:7071The browser walks through GitHub device-code sign-in when no compatible credential is already available.
Want persistent memory, Azure Functions, and a path to Copilot Studio? Skip the brainstem and go straight to Tier 2:
Mac / Linux:
curl -fsSL https://github.com/microsoft/aibast-agents-library/releases/download/installers/community_rapp-install.sh | bashWindows:
irm https://github.com/microsoft/aibast-agents-library/releases/download/installers/community_rapp-install.ps1 | iexCreates ~/rapp-projects/my-project/ β isolated project with its own venv, agents, and local storage. Auth happens through the chat UI (GitHub device code flow). No API keys needed to start.
Hippocampus setup and API guide | Installer source
The brainstem is a Flask server that connects to GitHub Copilot's API for LLM inference. You define a soul (system prompt) and drop in agents (Python tools the LLM can call). That's it.
~/.brainstem/src/rapp_brainstem/
βββ brainstem.py # the server
βββ soul.md # personality (system prompt)
βββ agents/ # auto-discovered tools
β βββ hello_agent.py
βββ local_storage.py # local-first storage shim
βββ .env # config (model, paths, port)
Any *_agent.py file in your agents directory gets auto-discovered and registered as a tool:
from basic_agent import BasicAgent
class WeatherAgent(BasicAgent):
def __init__(self):
self.name = "Weather"
self.metadata = {
"name": self.name,
"description": "Gets the weather for a city.",
"parameters": {
"type": "object",
"properties": {
"city": {"type": "string", "description": "City name"}
},
"required": ["city"]
}
}
super().__init__()
def perform(self, city="", **kwargs):
return f"It's sunny in {city}!"The Microsoft AI Academy is an experimental learning surface for 72 certified workshops and their packaged, reusable skills. Progress stays private in local browser storage by default; learners may optionally publish GitHub-attributed proof. The workshop path starts in GitHub Copilot and the local RAPP Brainstem, then carries the same skills into Copilot Studio, Teams, and Microsoft 365 Copilot. Academy completion is not an official credential or certification, and the Academy is not a replacement for Microsoft Learn.
The Agent Library is the browsable catalog: search across every industry template, filter by vertical, and copy the install command for one agent or a whole stack. Industry groups have shareable deep links, such as Financial Services, so recipients land on the preselected group instead of choosing it manually. The library reads registry.json, which build_registry.py regenerates from the agent manifests on every push.
The files themselves live under agents/@aibast-agents-library/, grouped into *_stacks/*_stack/ folders β a stack is the set of agents that solves one industry scenario end to end. Every template is a shell: the agent's shape for one business job, which wraps around your own live process and never copies your data (the AIBAST RAPP pattern). Review a template, adapt it to your environment, then drag the trusted *_agent.py file into the Brainstem chat or place it in the configured agents directory. Agent files are Python code and execute locally, so review them before installation.
Workshop achievements record self-paced milestones locally and let learners explicitly sync canonical badges to a public GitHub-attributed profile. The profile page shows the learner's achievement score.
Library metrics publishes AIBAST-only public downloads, every-file observations, per-agent rating Discussion upvotes, signed-in acquisition Discussion signals, workshop adoption, and opt-in achievement rollups. Each run also generates a shareable weekly/monthly PDF plus email-ready text under reports/. scripts/build_metrics.py builds the snapshot, and scripts/build_impact_report.py turns its dated history into the exportable impact report without converting unavailable coverage to zero. Visitor behavior on the published site (heatmaps, session recordings) comes from Microsoft Clarity; the tag is stamped on every public page from clarity.json by scripts/apply_clarity_tag.py (setup in docs/CLARITY.md).
RAPP teaches you the Microsoft AI stack one layer at a time. Start with the brainstem locally, then layer up when you're ready.
The survival basics. The brainstem runs the core agent loop β soul, tool-calling, conversation. Your GitHub Copilot subscription is the AI engine.
What you learn: Python agents, function-calling, prompt engineering, local-first development.
Give your brainstem a cloud body. Deploy to Azure so it's always-on with persistent storage, monitoring, and Azure OpenAI.
# Deploy via script
curl -fsSL https://raw.githubusercontent.com/microsoft/aibast-agents-library/main/deploy.sh | bashCreates: Function App (Python 3.11), Azure OpenAI (GPT-4o), Storage Account, Application Insights. All Entra ID auth β no API keys.
What you learn: ARM templates, Azure Functions, managed identity, RBAC, Azure OpenAI.
Connect your agent to Teams and M365 Copilot. Import the included Power Platform solution (MSFTAIBASMultiAgentCopilot_*.zip) into Copilot Studio, point it at your Azure Function, and publish.
The same agent logic you tested locally now answers in Microsoft Teams and M365 Copilot across your organization.
What you learn: Copilot Studio, declarative agents, Power Platform solutions, Teams integration, enterprise AI.
All config via .env (see .env.example):
| Variable | Default | Description |
|---|---|---|
GITHUB_TOKEN |
auto-detected via gh |
GitHub PAT or Copilot token |
GITHUB_MODEL |
auto |
Selects the fastest available Claude Haiku, then Sonnet, then gpt-4o; a model selected in the UI overrides this value. |
SOUL_PATH |
./soul.md |
Path to your soul file |
AGENTS_PATH |
./agents |
Path to your agents directory |
PORT |
7071 |
Server port |
BRAINSTEM_LAN_MODE |
false |
Opt in to LAN binding; non-loopback capability routes require the per-install secret. |
BRAINSTEM_ALLOWED_HOSTS |
(empty) | Optional comma-separated LAN hostnames. |
VOICE_ZIP_PASSWORD |
(empty) | Optional password for encrypted Azure Speech or ElevenLabs configuration. |
| Endpoint | Method | Description |
|---|---|---|
/chat |
POST | {"user_input": "...", "conversation_history": [], "session_id": "..."} |
/chat/stream |
POST | Server-sent event stream for chat and agent activity. |
/health |
GET | Status, model, loaded agents, token state |
/login |
POST | Start GitHub device code OAuth flow |
/models |
GET | List available models |
/agents |
GET | List installed agent files and loaded tools. |
/diagnostics/report |
POST | Prepare a privacy-scrubbed GitHub issue draft for review. |
- Python 3.11+
- Git
- GitHub account with Copilot access
cd ~/.brainstem/src && git pullrm -rf ~/.brainstem ~/.local/bin/brainstemThis project has adopted the Microsoft Open Source Code of Conduct.
This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft's Trademark & Brand Guidelines.
MIT License β Copyright (c) 2026 Microsoft.