corpus

Corpus

A "good enough" and easy to deploy RAG service.

⚠️ Disclaimer

Corpus is currently under active development and should be considered a work in progress. The API is in a preliminary stage and may not be stable. Please be aware that changes, including modifications, updates, or deprecations, can occur at any time without prior notice.

Features

• OIDC authentication with email-based role mapping and access whitelist
• Markdown-based chunking
• Use full-text and vector-based indexes (via Bleve and SQLite Vec)
• Web interface and REST API
• Backup and restore via the REST API
• CLI with abstract filesystem watching and auto-indexing (local, S3, FTP, SFTP, WebDAV, SMB...)

Getting started

With Docker

# Create data volume
docker volume create corpus_data

# Start container
docker run \
  -it --rm \
  -v corpus_data:/data \
  --net=host \
  -e CORPUS_LLM_PROVIDER_KEY="<LLM_SERVICE_API_KEY>" \
  -e CORPUS_LLM_PROVIDER_BASE_URL="<LLM_SERVICE_BASE_URL>" \
  -e CORPUS_LLM_PROVIDER_CHAT_COMPLETION_MODEL="<LLM_SERVICE_CHAT_COMPLETION_MODEL>" \
  -e CORPUS_LLM_PROVIDER_EMBEDDINGS_MODEL="<LLM_SERVICE_EMBEDDINGS_MODEL>" \
  -e CORPUS_HTTP_AUTHN_PROVIDERS_GITHUB_KEY="<github_oauth2_app_key>"
  -e CORPUS_HTTP_AUTHN_PROVIDERS_GITHUB_SECRET="<github_oauth2_app_secret>" \
  -e CORPUS_HTTP_AUTHN_PROVIDERS_GITHUB_SCOPES=openid,user \
  ghcr.io/bornholm/corpus-server:latest

Then open http://localhost:3002 in your browser.

Examples

With a local ollama instance:

CORPUS_LLM_PROVIDER_BASE_URL=http://127.0.0.1:11434/v1/
CORPUS_LLM_PROVIDER_CHAT_COMPLETION_MODEL=qwen2.5:7b
CORPUS_LLM_PROVIDER_EMBEDDINGS_MODEL=mxbai-embed-large

With Mistral:

CORPUS_LLM_PROVIDER_BASE_URL=https://api.mistral.ai/v1/
CORPUS_LLM_PROVIDER_CHAT_COMPLETION_MODEL=mistral-small-latest
CORPUS_LLM_PROVIDER_EMBEDDINGS_MODEL=mistral-embed

Sponsors

License

AGPL-3.0