mudler/LocalRecall

Your AI. Your Hardware. Your Rules.

A lightweight, no-frills RESTful API designed for managing knowledge bases and files stored in vector databases—no GPU, internet, or cloud services required! LocalRecall provides a simple and generic abstraction layer to handle knowledge retrieval, ideal for AI agents and chatbots to manage both long-term and short-term memory seamlessly.

Currently, LocalRecall is batteries included and supports multiple vector database engines:

• Chromem: Local file-based vector store (default)
• PostgreSQL: Production-ready PostgreSQL with TimescaleDB, pgvector, and pgvectorscale for hybrid search (BM25 + vector similarity)

It can easily integrate with LocalAI, LocalAGI, and other agent frameworks, offering an intuitive web UI for convenient file management, including support for raw text inputs.

📚🆕 Local Stack Family

🆕 LocalAI is now part of a comprehensive suite of AI tools designed to work together:

---

🌟 Features

• ⚡ RESTful API: Simple and intuitive REST interface for knowledge management.
• 📡 Fully Local: Operates offline without external cloud dependencies.
• 📚 RAG Knowledgebase: Retrieve-Augmented Generation (RAG) compatible with multiple vector databases.
• 🗃️ Memory Management: Ideal for AI-driven applications requiring memory abstraction.
• 📂 File Support:
  • ✅ Markdown
  • ✅ Plain Text
  • ✅ PDF
  • ⏳ More formats coming soon!

---

⚙️ Prerequisites

• Go 1.16 or higher
• Docker (optional, for containerized deployment)

---

🚧 Quickstart

📥 Clone Repository

git clone https://github.com/mudler/LocalRecall.git
cd LocalRecall

🛠️ Build from Source

go build -o localrecall

▶️ Run Application

./localrecall

Your web UI will be available at http://localhost:8080.

---

🐳 Docker Deployment

Using Chromem (Default)

Build and run using Docker:

docker build -t localrecall .
docker run -ti -v $PWD/state:/state \
               -e COLLECTION_DB_PATH=/state/db \
               -e EMBEDDING_MODEL=granite-embedding-107m-multilingual \
               -e FILE_ASSETS=/state/assets \
               -e OPENAI_API_KEY=sk-1234567890 \
               -e OPENAI_BASE_URL=http://localai:8080 \
               -p 8080:8080 localrecall

# Or use the images already built by the CI:
docker run -ti -v $PWD/state:/state \
               -e COLLECTION_DB_PATH=/state/db \
               -e EMBEDDING_MODEL=granite-embedding-107m-multilingual \
               -e FILE_ASSETS=/state/assets \
               -e OPENAI_API_KEY=sk-1234567890 \
               -e OPENAI_BASE_URL=http://localai:8080 \
               -p 8080:8080 quay.io/mudler/localrecall

Using PostgreSQL (Recommended for Production)

For production deployments, PostgreSQL provides better performance, scalability, and hybrid search capabilities (combining BM25 keyword search with vector similarity search).

Quick Start with Docker Compose

The easiest way to get started with PostgreSQL is using Docker Compose:

docker compose up -d

This will start:

• LocalAI: For embeddings (port 8081)
• PostgreSQL: With TimescaleDB, pgvector, and pgvectorscale extensions (port 5432)
• LocalRecall: RAG server configured to use PostgreSQL (port 8080)

Manual Setup

1. Start PostgreSQL (using the pre-built image):

docker run -d \
  --name localrecall-postgres \
  -e POSTGRES_DB=localrecall \
  -e POSTGRES_USER=localrecall \
  -e POSTGRES_PASSWORD=localrecall \
  -p 5432:5432 \
  -v postgres_data:/var/lib/postgresql/data \
  quay.io/mudler/localrecall:latest-postgresql

2. Start LocalRecall with PostgreSQL:

docker run -ti \
  -e DATABASE_URL=postgresql://localrecall:localrecall@localhost:5432/localrecall?sslmode=disable \
  -e VECTOR_ENGINE=postgres \
  -e EMBEDDING_MODEL=granite-embedding-107m-multilingual \
  -e FILE_ASSETS=/assets \
  -e OPENAI_API_KEY=sk-1234567890 \
  -e OPENAI_BASE_URL=http://localai:8080 \
  -e HYBRID_SEARCH_BM25_WEIGHT=0.5 \
  -e HYBRID_SEARCH_VECTOR_WEIGHT=0.5 \
  -p 8080:8080 \
  quay.io/mudler/localrecall

PostgreSQL Features

• Hybrid Search: Combines BM25 (keyword) and vector (semantic) search with configurable weights
• Advanced Indexing:
  • GIN indexes for full-text search
  • BM25 indexes for keyword search
  • DiskANN/HNSW indexes for vector similarity search
• Extensions Included:
  • pg_textsearch: BM25 keyword search
  • vectorscale: Advanced vector search with DiskANN
  • pgvector: Vector similarity search (fallback)
  • timescaledb: Time-series capabilities

---

🌍 Environment Variables

LocalRecall uses environment variables to configure its behavior. These variables allow you to customize paths, models, and integration settings without modifying the code.

These variables can be passed directly when running the binary or inside your Docker container for easy configuration.

You can use an .env file to set the variables. The Docker compose file is configured to use an .env file in the root of the project if available.

---

📖 REST API

Base URL: http://localhost:8080/api

🔧 Manage Collections

• Create Collection:

curl -X POST $BASE_URL/collections \
  -H "Content-Type: application/json" \
  -d '{"name":"myCollection"}'

• Upload File:

curl -X POST $BASE_URL/collections/myCollection/upload \
  -F "file=@/path/to/file.txt"

• List Collections:

curl -X GET $BASE_URL/collections

• List Files in Collection:

curl -X GET $BASE_URL/collections/myCollection/entries

• Get Entry Content:

curl -X GET $BASE_URL/collections/myCollection/entries/file.txt

Returns collection, entry, chunks (array of id, content, metadata), and count.

• Get Entry Raw File:

curl -X GET $BASE_URL/collections/myCollection/entries/file.pdf/raw

Returns the original uploaded binary file with the appropriate Content-Type header.

• Search Collection:

curl -X POST $BASE_URL/collections/myCollection/search \
  -H "Content-Type: application/json" \
  -d '{"query":"search term", "max_results":5}'

• Reset Collection:

curl -X POST $BASE_URL/collections/myCollection/reset

• Delete Entry:

curl -X DELETE $BASE_URL/collections/myCollection/entry/delete \
  -H "Content-Type: application/json" \
  -d '{"entry":"file.txt"}'

• Add External Source:

curl -X POST $BASE_URL/collections/myCollection/sources \
  -H "Content-Type: application/json" \
  -d '{"url":"https://example.com", "update_interval":30}'

The update_interval is specified in minutes. If not provided, it defaults to 60 minutes.

External sources support various URL types:

• Web pages (https://example.com)
• Git repositories (https://github.com/user/repo.git or [email protected]:user/repo.git)
• Sitemaps (https://example.com/sitemap.xml)

For private Git repositories, set the GIT_PRIVATE_KEY environment variable with a base64-encoded SSH private key:

# Encode your private key
export GIT_PRIVATE_KEY=$(cat /path/to/private_key | base64 -w 0)

• Remove External Source:

curl -X DELETE $BASE_URL/collections/myCollection/sources \
  -H "Content-Type: application/json" \
  -d '{"url":"https://example.com"}'

External sources are automatically monitored and updated in the background. The content is periodically fetched and added to the collection, making it searchable through the regular search endpoint.

---

🔌 Model Context Protocol (MCP) Integration

LocalRecall can be controlled via MCP (Model Context Protocol) through the LocalRecall MCP Server available in the mcps repository.

The MCP server provides tools for:

• 🔍 Search: Search content in collections
• ➕ Create Collection: Create new collections
• 🔄 Reset Collection: Clear collections
• 📄 Add Document: Upload documents to collections
• 📋 List Collections: List all available collections
• 📁 List Files: List files in a collection
• 🗑️ Delete Entry: Remove entries from collections

Quick Start with MCP

The MCP server can be configured to enable specific tools for security and flexibility:

docker run -e LOCALRECALL_URL=http://localhost:8080 \
           -e LOCALRECALL_API_KEY=your-api-key \
           -e LOCALRECALL_ENABLED_TOOLS="search,list_collections,add_document" \
           ghcr.io/mudler/mcps/localrecall:latest

For more information and configuration options, see the LocalRecall MCP Server documentation.

---

📝 License

Released under the MIT License.

---

🤝 Contributing

We welcome contributions! Please feel free to:

• ✅ Open an issue for suggestions or bugs.
• ✅ Submit a pull request with enhancements.