Agent-Me v2.0.0 provides a runnable open-source reference implementation for inspecting grounded, role-based multi-agent workflows, paired with a bilingual hands-on engineering curriculum.
Reference implementation highlights
- Planner → Researcher → Critic → Writer orchestration with immutable typed handoffs
- Deterministic retrieval over a reviewable Markdown knowledge base
- Evidence-aware answers, critic-gated abstention, and safe operational traces
- Repeatable behavioral evaluation for supported and unsupported requests
- FastAPI + React reference stack with contract tests and Docker Compose integration in CI
- Local extractive and collaboration paths that do not require a paid model API
The agents are sequential, in-process role stages. This release does not claim a distributed runtime, autonomous swarm, general-purpose agent SDK, factual-correctness guarantee, or production-ready enterprise platform.
Engineering curriculum
The eight numbered lessons explain and rebuild the same architecture used by the reference implementation. Complete learning paths are available in English and Simplified Chinese, with a rubric, glossary, contribution workflow, and reproducible labs.
Start here
- Run the reference implementation.
- Inspect the architecture.
- Follow the engineering learning path.
- Ask questions in Discussions Q&A.
Upgrade note
This release moves beyond the original starter demo by combining a tested reference implementation with a structured bilingual curriculum. Existing API routes remain available.
Full changes: v1.1.0...v2.0.0