corpus

Corpus

An 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

Usage as a library

Corpus can be embedded directly in a Go project to index and search documents without running a server.

Installation

go get github.com/bornholm/corpus

Note: The sqlitevec adapter uses CGO. Make sure a C compiler is available in your build environment.

Quick start

import (
    "context"

    "github.com/bornholm/corpus/pkg/corpus"
    "github.com/bornholm/genai/llm/provider"
    providerenv "github.com/bornholm/genai/llm/provider/env"
)

// Create a client from environment variables (see configuration below)
llmClient, _ := provider.Create(ctx, providerenv.With("LLM_", ".env"))

// Initialise an embedded corpus (SQLite + Bleve + SQLiteVec, auto-composed)
c, err := corpus.New(ctx,
    corpus.WithStoragePath("./data"),
    corpus.WithLLMClient(llmClient),
)

// Create a collection, index a file, search, ask
collID, _ := c.CreateCollection(ctx, "my-docs")
taskID, _ := c.IndexFile(ctx, collID, "notes.md", reader)

results, _ := c.Search(ctx, "my question",
    corpus.WithSearchCollections(collID),
    corpus.WithSearchMaxResults(5),
)

answer, _, _ := c.Ask(ctx, "my question", results)

LLM configuration

The providerenv.With(prefix) helper reads LLM configuration from environment variables:

Variables can also be loaded from a .env file by passing its path to providerenv.With.

More advanced options are available. A full runnable example is available in example/embedded/.

Sponsors

License

AGPL-3.0