Repolode
Submit
jzjzzzzzzz

jzjzzzzzzz/agent-me

Open-source, inspectable framework for building an AI Agent Twin that distills your knowledge, memories and decisions into a grounded personal assistant.

View on GitHubWebsite
Stars
586
Forks
54
Open issues
1
Last commit
yesterday

What is it?

Agent-Me is a FastAPI and React project for building a personal AI assistant from reviewable Markdown knowledge. Instead of sending a question straight to a model, it runs a pipeline: retrieval over your own documents, a planner, a researcher that collects exact source excerpts, a critic that blocks answers without enough evidence, a writer that produces a citation-aware response, and an optional verifier that checks citation paths and output invariants.

Each run returns a public trace showing the role outcomes, safe intermediate summaries, evidence and metrics, so the reasoning path can be inspected. The project does not expose a model's private chain of thought. Retrieval is deterministic and local by default, and an optional OpenAI-compatible provider can be configured for chat answers. An opt-in private workspace adds SQLite-backed memory, persisted chat transcripts and a browser review workbench.

The public repository uses fictional example data, and the project states that the verifier checks structure and citations rather than factual truth.

Key features

  • 01Answers grounded in your Markdown knowledge with exact source excerpts
  • 02Sequential Planner, Researcher, Critic, Writer and optional Verifier roles
  • 03Public execution trace with role outcomes and metrics
  • 04Deterministic local retrieval, with optional OpenAI-compatible provider mode
  • 05Opt-in private workspace with SQLite memory and a browser review workbench
  • 06Versioned evaluation cases run in CI

Use cases

  • Building a personal assistant that answers questions from your own notes and documents
  • Studying how multi-agent pipelines with evidence gates are structured
  • Prototyping an inspectable RAG workflow that refuses to answer without sources
  • Exploring memory and provenance design for personal AI systems

Topics

#ai-agent#rag#llm#personal-ai#multi-agent#knowledge-base#fastapi#agentic-ai#ai#ai-agents

FAQ

What is agent-me?
Open-source, inspectable framework for building an AI Agent Twin that distills your knowledge, memories and decisions into a grounded personal assistant. Agent-Me is a FastAPI and React project for building a personal AI assistant from reviewable Markdown knowledge. Instead of sending a question straight to a model, it runs a pipeline: retrieval over your own documents, a planner, a researcher that collects exact source excerpts, a critic that blocks answers without enough evidence, a writer that produces a citation-aware response, and an optional verifier that checks citation paths and output invariants.
Is agent-me free to use? What license does it have?
agent-me is open source under the MIT license. Check the license terms before using it in commercial products.
Is agent-me still maintained?
Actively maintained. The last commit to agent-me was on Oct 21, 2026.
What can I use agent-me for?
Building a personal assistant that answers questions from your own notes and documents Studying how multi-agent pipelines with evidence gates are structured Prototyping an inspectable RAG workflow that refuses to answer without sources Exploring memory and provenance design for personal AI systems
Where can I get agent-me?
The source code and installation instructions are on GitHub: https://github.com/jzjzzzzzzz/agent-me

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects