TypeAgent
Microsoft Research sample code exploring a single personal agent with natural-language interfaces.
About
TypeAgent is Microsoft's research exploration of what a single personal agent should look like: one agent with a natural-language interface that routes requests to specialized application agents for calendar, email, music, lists, and the browser, while distilling model behavior into logical structures to keep cost and latency low. Actions and memories flow together, so a calendar entry you add today becomes a parameter for a request tomorrow.
The project is working toward an AMP architecture (actions, memory, plans) and uses Structured RAG, a new indexing approach for agent memory that answers questions about past conversations substantially better than classic RAG. Implementations exist in TypeScript and Python, with an agent dispatcher, shell example, and VS Code integration.
Be clear about what this is: the README calls it sample code, not a product or supported framework. There are no published installable builds; you clone and build from the repo. Valuable as architecture reference and research, not as something to ship on.
What it can do
- Hub-and-spoke personal agent with application-agent plugins
- Structured RAG for conversational memory
- AMP architecture exploration (actions, memory, plans)
- TypeScript and Python implementations
- MIT-licensed research sample code
Setup
Clone microsoft/typeagent; install Node 22+ and pnpm; run pnpm i && pnpm run build in ts/, then run the shell example. Python port at microsoft/typeagent-py.
Prerequisites
Node 22+ and pnpm (TypeScript) or Python (typeagent-py); an LLM provider key; research mindset.
Use cases
Works with
Trust score — 6/10
Scored 6/10 (verified tier, metrics-v1). Evidence: 748 GitHub stars (microsoft/typeagent); repo pushed 2026-10-15 (same-day activity); MIT license confirmed via GitHub API; Dispatcher architecture, Structured RAG, and AMP direction verified against official README and docs. Manual review: explicitly sample code, not a supported framework -- no installable builds; value is architectural reference, not something to ship on. License: MIT.
★ 748 GitHub stars
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 4.0/10
- Maintenance · recency, releases, issue load
- 8.0/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 6.0/10
- · 748 GitHub stars (microsoft/typeagent); repo pushed 2026-10-15 (same-day activity); MIT license confirmed via GitHub API
- · Dispatcher architecture, Structured RAG, and AMP direction verified against official README and docs
- · manual review: explicitly sample code, not a supported framework -- no installable builds; value is architectural reference, not something to ship on
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.