Skip to main content
SpeakerWeave ships two production-ready deployment paths in the repository. Both are exercised against the live site.

​
Railway (reference)

Two services from the same repo:
  1. API — root directory api, start uvicorn main:app --host 0.0.0.0 --port $PORT, health check /health.
  2. Web — root directory web, builds the included Dockerfile (nginx serving the Vite build, precompressed assets, and the backend proxy). Set BACKEND_URL to the API origin.
Cross-service URLs (FRONTEND_URL, PUBLIC_APP_URL, PUBLIC_API_URL, CORS_ALLOWED_ORIGINS) are documented in api/.env.example.

​
Cloudflare Workers (edge web tier)

The web tier also ships as a Cloudflare Worker — static assets from the edge with Brotli and SPA fallback, and the identical /api, /public, /mcp, and OAuth proxy contract, including SSE chat streaming:
Configuration lives in web/wrangler.jsonc (point BACKEND_URL at your API origin). The reference edge deployment runs live at speakerweave-web.brandon-c2f.workers.dev against the same API and database as the primary site.
The API tier is a standard Python/FastAPI container — Railway is the reference home, and any container host (Fly, Render, ECS, a VPS) works unchanged. Run migrations before code that requires them.

​
Anything else

The hosting contract is deliberately small: serve the SPA with history fallback, route /api, /public, /mcp, /oauth, and /.well-known/oauth-* to the API origin, keep secrets server-side, and inject VITE_* at build time. AGENTS.md documents the invariants for adapting either tier.