Colibri

Colibri — Tokenomics Trifecta (Rust + Go) — Work in Progress

Control plane for autonomous AI agents with tokenomics trifecta: performance / speed / cost. Target metric: cost-per-intelligence > cost-per-token.

⚠️ Note: Still in active development. Components work individually, but full chain (boot → seed → Pi agent → Colibri daemon → Mother/Hive → benchmark) not end-to-end validated on hardware.

Technologies: Rust (core: daemon, CLI, zot) · Go (services: API, MCP bridge, cost dashboard) · PostgreSQL + pgvector · bhyve VMs · Bastille jails

Components: View status matrix in wiki →

• colibri-daemon (Rust) — core loop, systemd-style supervision ✅
• colibri-cli (Rust) — basics, missing advanced subcommands 🔄
• zot agent harness (Rust) — NDJSON RPC, subtask delegation, tool interception ✅
• colibri-mcp (Go) — SSH-stdio bridge, tool proxy ✅
• cost-dashboard (Go) — in development 🔄
• llama.cpp benchmarks (Rust/Go) — /v1 timings object for ms/tok ✅

Documentation: Colibri wiki — 44 pages (LLM benchmarks, architecture decisions, OKF v0.2, pre-push hooks)

Blockers for end-to-end: bhyve spawn logic in ISO 🔄 · pgvector migration script 🔄 · cost-dashboard integration 🔄 · hardware test on Valhala 📅