Docs home
Documentation
Everything from scratch to NPU, written to be read. The docs describe the engine as it is, not as it is pitched. Build it, run it, pick a model, talk to it.
Getting started
The one-line install, running your first model, and a health check. Shortest path from nothing to a model.
User · BuildBuilding the engine
TheRock 7.15.0a, the onebin target, FastFlowLM, optional GPU decode, and the exact gotchas that stop people.
User · ServeRunning the server
Weights, launch, health check, and the OpenAI-compatible API. Serve the same binary over HTTP.
User · VoiceJARVIS
Mic to VAD to STT to LLM to TTS to speaker, all in-process, pure C++, on any backend. The engine is the app.
Reference · CoverageModel families
16 families, the 1BP canonical format, and the Zyphra stack the engine was tuned against.
Reference · MeasuredBenchmarks
Kernel-level and end-to-end numbers, with honest labels and the gaps called out.
Community · Support
Ask the team
Official 1bit.MONSTER support live on Fluxer. Post a question, get an answer from the engine team.