brain

Overview ¶

Command brain runs the model-turn orchestration pool: it claims model_turn work from the shared Postgres queue, replays session event logs into provider requests, and writes the turns back as Anthropic-native events. Stateless and horizontally scalable — run as many as needed. Configuration is environment-driven:

DATABASE_URL          Postgres DSN (required; same database as the
                      controlplane)
MODEL_PROVIDERS_PATH  JSON file mapping model strings to provider
                      endpoints (required) — see internal/provider
BLOB_BACKEND + BLOB_*  object storage, as controlplane: "s3" (default when
                      empty) or "gcs". Read to grade file-output rubrics
                      against the bytes; without it they are graded from
                      the description alone
BRAIN_LEASE_TTL       work-item lease, Go duration (default "2m")
BRAIN_POLL_INTERVAL   idle queue poll, Go duration (default "250ms")
OTEL_EXPORTER_OTLP_ENDPOINT  optional OTLP/gRPC collector endpoint
OTEL_EXPORTER_OTLP_INSECURE  "true" to export without TLS (default TLS)