Configuration and providers

Configuration sources

• .env.example documents database, pgvector, Ollama, model, and machine-profile variables
• machine.toml.example describes machine-specific service and workspace behavior
• CLI flags can override connection and logging choices for a run

Model routing

Ollama provides the local-first path. The gateway can route to configured hosted providers such as OpenAI, Anthropic, and Fireworks while keeping provider choice outside agent logic.

Gateway behavior

The TypeScript gateway runs separately on port 18790. It performs local-first provider selection, supports streamed responses, tracks usage and cost, and uses circuit-breaker behavior to avoid repeatedly selecting an unhealthy provider.

Machine profiles

`machine.toml` captures host-specific GPU, workspace, service, and provider choices. The initializer detects hardware and writes practical defaults; explicit environment and command-line values can override them for automation.

Secret handling

Provider credentials belong in the runtime environment or protected configuration. They should not appear in prompts, task payloads, or logs.