Skip to content

configuration.md

Latest commit

 

History

History
92 lines (67 loc) · 3.83 KB

configuration.md

File metadata and controls

92 lines (67 loc) · 3.83 KB
Raw

Configuration

All configuration is via environment variables. Copy .env.example to .env and edit.

Environment

Variable Required Default Description
APP_ENV No development development, test, or production

Production mode emits JSON logs and rejects development-only or incomplete security configuration. Development defaults are never silently promoted to production behavior.

Database

Variable Required Default Description
DATABASE_URL Yes — Postgres connection string

Format: postgres://<user>:<password>@<host>:<port>/<database>

Auth

Variable Required Default Description
JWT_SECRET Yes — Secret key for signing JWT tokens

Rustwing does not provide a fallback JWT secret. Generate a strong, unique secret for every environment; the API fails at startup when it is absent. Production also rejects secrets shorter than 32 characters and known example placeholders. The value is not logged.

LLM

Variable Required Default Description
LLM_PROVIDER No stub Provider: "disabled", "stub", "deepseek", "openai", "gemini", or "anthropic"
LLM_MODEL No provider default Model name for the selected provider
DEEPSEEK_API_KEY For DeepSeek — API key from DeepSeek
OPENAI_API_KEY For OpenAI — API key from OpenAI
GEMINI_API_KEY For Gemini — API key from Google Gemini
ANTHROPIC_API_KEY For Anthropic — API key from Anthropic
LLM_MAX_TOKENS No — Default max output tokens (safety cap). Override per-request in code.

Set LLM_PROVIDER=stub for local development — no API key needed. The stub returns a canned response without usage metrics and does not log prompt contents. If LLM_MODEL is unset, Rustwing picks a provider-specific default: deepseek-chat, gpt-4o, gemini-2.5-flash, or claude-sonnet-4-6.

Per-request max_tokens override is supported in service code via LlmRequest { prompt, max_tokens: Some(512) }. When set, it takes precedence over the LLM_MAX_TOKENS env default.

The stub is rejected in production; use disabled when a production binary does not need LLM calls. Unknown providers, missing credentials, and provider initialization failures stop startup instead of falling back.

Logging

Variable Required Default Description
RUST_LOG No info,api=debug Log level and targets

Format: [target=]level[,target=level...]

Examples:

  • info — info and above for all crates
  • info,api=debug — debug for your app, info for dependencies
  • trace — everything (very verbose)

Worker

Variable Required Default Description
WORKER_TICK_SECONDS No 10 Seconds between worker processing ticks
WORKER_LEASE_SECONDS No 60 Job lease duration; active jobs heartbeat at one third of this interval
WORKER_BATCH_SIZE No 10 Jobs claimed per polling batch, capped at 100
WORKER_ID No Generated UUIDv7 name Stable diagnostic identity for this process

The worker uses the same DATABASE_URL, LLM_PROVIDER, LLM_MODEL, and provider API keys as the API binary.

Rate limiting

Variable Required Default Description
RATE_LIMIT_GLOBAL_REQUESTS No 600 Requests per process-local window
RATE_LIMIT_AUTH_REQUESTS No 10 Login/registration requests per process-local window
RATE_LIMIT_WINDOW_SECONDS No 60 Window duration

The limiter is per process and is not a globally distributed rate limiter. It keys requests by the direct peer IP. Deployments behind a proxy or with multiple replicas should enforce the authoritative client policy at a trusted edge rather than trusting arbitrary forwarded headers.