Docker¶

New in 0.3.1. Every release publishes a Docker image of the HTTP API to the GitHub Container Registry:

ghcr.io/sadrmanesh/persian-rag

Tag	Points to
`0.3.1`	That exact release
`0.3`	The newest 0.3.x release
`latest`	The newest release

Images are built for linux/amd64 and linux/arm64. They contain persian-rag[server,pdf,docx] on Python 3.12, run as a non-root user (uid 10001), and listen on port 8000. The working directory and volume is /data, and the default command serves the index in /data/index.

Index and serve¶

Build an index from a folder of documents into a named volume, then serve it:

docker volume create persian-rag-data

docker run --rm \
  -v "$PWD/library:/library:ro" \
  -v persian-rag-data:/data \
  ghcr.io/sadrmanesh/persian-rag:0.3.1 ingest /library --index /data/index

docker run -d --name persian-rag -p 8000:8000 \
  -v persian-rag-data:/data \
  ghcr.io/sadrmanesh/persian-rag:0.3.1

curl -s localhost:8000/health
curl -s localhost:8000/retrieve -H 'content-type: application/json' \
  -d '{"query": "حافظ اهل کجا بود؟", "k": 3}'

The entrypoint is the persian-rag command, so any subcommand works: ingest, query, chat, eval, or serve with your own flags. An index you built on the host can be mounted directly with -v "$PWD/.persian-rag:/data/index:ro".

Answers with a language model¶

The image includes the OpenAI-compatible and Ollama adapters. Pass the key as an environment variable and the flags to serve:

docker run -d -p 8000:8000 -v persian-rag-data:/data \
  -e OPENAI_API_KEY \
  ghcr.io/sadrmanesh/persian-rag:0.3.1 \
  serve --index /data/index --host 0.0.0.0 --port 8000 --llm openai --citations

For Ollama on the host, add -e OLLAMA_HOST=http://host.docker.internal:11434 (and --add-host=host.docker.internal:host-gateway on Linux) and use --llm ollama.

Health check¶

The image has a Docker HEALTHCHECK that calls GET /health every 30 seconds. docker ps shows the container as healthy once the index is loaded. If you change the port in the command, also set -e PERSIAN_RAG_PORT=<port> so the check uses it.

Building the image yourself¶

docker build -t persian-rag .
docker build -t persian-rag --build-arg EXTRAS=server,pdf,docx,sentence-transformers .

EXTRAS selects the extras installed in the image. The published image leaves out sentence-transformers, which adds a few gigabytes; build your own image with it to serve an index made with --embedder sentence-transformers.

The server has no authentication. Put it behind your own gateway before exposing it.