henomis / phero

What is phero?

Phero is a modern Go framework for building multi-agent AI systems, where agents collaborate like ants toward shared goals via orchestration, tool calling, and shared memory. It solves the pain of wiring LLMs, embeddings, RAG pipelines, and vector stores (Qdrant or Postgres pgvector) from scratch, delivering conversational agents, multi-turn REPLs, and workflows like plan-execute-critique in minimal code. OpenAI-compatible, it runs local with Ollama and exposes Go functions as tools with auto-JSON schemas.

Why is it gaining traction?

Its composable, lightweight design beats bloated alternatives by prioritizing interfaces for swapping LLMs or stores, plus built-in skills from markdown files and MCP tool integration. Developers hook on the ant-inspired philosophy—simple primitives for complex coordination—and battle-tested examples like debate committees or RAG chatbots that just work. In Go's efficient world, it feels like modern frameworks for web development but tuned for AI agents.

Who should use this?

Backend Go devs prototyping multi-agent automations, like research pipelines or customer support bots with specialized roles (judge, worker). Suited for teams needing RAG over docs or long-term memory without Python overhead—think CLI tools handling file ops, code execution, or human-in-loop decisions. Avoid if you're locked into JS ecosystems or need enterprise-scale production yet.

Verdict

Solid starter for Go AI experiments with excellent docs and examples; at 11 stars and 1.0% credibility, it's immature but passes lint/tests cleanly. Dive in via the simple agent demo if multi-agent appeals—contribute to accelerate it.

(187 words)