Manual setup

Use manual setup when you want to run OrcAI directly on a host instead of using the full Compose stack.

PrerequisitesCopy Anchor Link

• Bun >=1.3.10
• PostgreSQL
• Qdrant
• SpiceDB
• S3-compatible object storage
• An OpenAI-compatible endpoint for embeddings
• Tesseract with eng and deu language packs if workers run on the host
• An SMTP server (optional)

Install dependenciesCopy Anchor Link

cp .env.example .env
bun install --frozen-lockfile

Edit .env before starting the app. Configuration lists the required variables.

Initialise data storesCopy Anchor Link

Apply the database migrations and the SpiceDB schema:

bun run stack migrate

Package scripts do not read the root .env, so run them through bun run stack, which loads it. bun run stack exec -- <command> does the same for any other command.

Start development processesCopy Anchor Link

Start the app:

bun run dev

Start background workers in a separate terminal:

bun run workers:dev

The app is then at http://localhost:3000. On a fresh database, complete the instance setup at http://localhost:3000/init, as described in Initialise the instance.

Production commandsCopy Anchor Link

Build and start the app:

bun run build
bun run stack exec -- bun run start

Start workers:

bun run stack exec -- bun run workers:start

OCR on the hostCopy Anchor Link

OCR-backed asset preparation runs in the worker process. When workers run outside Docker, install Tesseract and the configured language packs. Use KREUZBERG_OCR_LANGUAGE=eng+deu only where both packs are installed.