Language models and failover
The brain uses a configured language-model provider path and tool-calling loop.
Configuration
Choose endpoints, models and fallbacks through Settings. A compatible protocol does not ensure equal tool-calling behavior or latency across providers.
Readiness requires a real authenticated answer. A model name or measured CLI flag does not prove working provider access.
Reliability
Inspect traces, errors and provider observations. Failover cannot remove quotas, network outages or differences in supported tools.
Implementation sources
Repository paths below describe the implementation. They do not certify a running deployment or a configured external account.
src/mushegh/brain/llm.pysrc/mushegh/brain/agent.pysrc/mushegh/config.py
Subsystem directory · HUD guide · Verification and known limits