We build and release research-grade developer tooling for the AI-native era — agents, proxies, councils and document systems — shipped open to the community.
Lucidity Labs is an independent open source R&D collective building the next generation of developer tools for AI-native engineering — proxies, councils, sub-agent fabrics, and the unglamorous infrastructure that makes them work.
Every project is shipped publicly on GitHub, with no gatekeeping, no lock-in and no waitlist — just code, in the open, for builders who want to move at the pace of the field.
Multi-provider Claude Code proxy
A Claude Code proxy that lets you run multiple model providers from a single, ergonomic VS Code panel — and a CLI for terminal-first workflows.
Karpathy's LLM-Council, retooled for developers
A developer-grade implementation of the LLM-Council concept — multiple models deliberating on code, architecture and design — exposed via MCP so agents can consult the council directly.
A custom harness for xAI's agentic swarm
Grok 4.20 Multi-Agent is a frontier model that runs an agentic swarm behind a single API — 4 agents in low-thinking mode, 16 in high. It's novel enough that most agent harnesses break on it, so we built one that doesn't, and wired it into Claude Code as a sub-agent, reviewer, or analyst — without leaving the editor.
Document management for software projects
A focused document management layer for engineering projects — specs, ADRs, prompts and runbooks, organized for humans and indexed for agents.
We're a small team. Real messages get real replies. Tell us what you're building and where we might fit.