Local models, open weights
Running real inference on hardware you control. Quantization, serving, fine-tunes, and the configs that took a weekend to get right.
A working group for builders who think intelligence should stay open. Open weights, local models, open agentic systems. We trade the beta that actually works — not the demos.
Request AccessThe intelligence stack is consolidating fast. This is a room for the people building the other path — and sharing how, in the open.
Running real inference on hardware you control. Quantization, serving, fine-tunes, and the configs that took a weekend to get right.
Open harnesses, runtimes, and the scaffolding that turns a model into something that does work. What holds up under real load.
Skills, MCP servers, evals, and agent tooling — built in the open, reviewed by people who actually run them in production.
End-to-end: retrieval, memory, orchestration, observability. Contributing upstream instead of rebuilding the same thing privately.
This is a small room, kept small on purpose. Access is reviewed by hand.
One paragraph. What you're working on, and what you'd bring to the room.
A human reads every application. You'll hear back either way.
The community runs on a self-hosted Buzz relay. We'll send install steps and you'll send back your public key.
Your key goes on the allowlist. Point the app at our relay and start building.
No pubkey needed yet — we'll walk you through that if you're in.