Berry¶

A semantic memory system for AI assistants

Berry provides persistent, searchable memory storage that bridges the gap between you and your AI tooling. Store information, questions, and requests, then retrieve them using natural language search powered by vector embeddings.

Why Berry?¶

AI assistants are powerful but stateless. Each conversation starts fresh, losing valuable context from previous sessions. Berry solves this by providing:

• Persistent Memory: Store important information that survives between sessions
• Semantic Search: Find relevant memories using natural language queries
• Structured Data: Categorize memories as questions, requests, or information
• Access Control: Control visibility with private, shared, and public memories
• AI-Native: Built-in MCP server for seamless AI assistant integration

Quick Start¶

Install¶

curl -fsSL https://raw.githubusercontent.com/geoffjay/berry-rs/main/scripts/install.sh | bash

Configure¶

# Initialize configuration
berry init

# That's it! Berry uses LanceDB by default (embedded, no setup needed).

# Optionally, to use ChromaDB instead:
# docker run -d -p 8000:8000 chromadb/chroma
# export BERRY_STORE=chroma

Start the Server¶

berry serve --foreground

Store and Search¶

# Store a memory
berry remember "The API uses JWT tokens for authentication"

# Search memories
berry search "authentication"

# Store with metadata
berry remember "How do I reset a password?" --type question --tags "auth,faq"

Architecture¶

┌─────────────────────────────────────────────────────────────┐
│                        Clients                               │
├─────────────┬─────────────┬─────────────┬───────────────────┤
│  berry CLI  │  MCP Server │  HTTP API   │  Your Application │
└──────┬──────┴──────┬──────┴──────┬──────┴─────────┬─────────┘
       │             │             │                │
       └─────────────┴──────┬──────┴────────────────┘
                            │
                    ┌───────▼───────┐
                    │ Berry Server  │
                    │  (HTTP API)   │
                    └───────┬───────┘
                            │
              ┌─────────────┼─────────────┐
              │             │             │
      ┌───────▼───────┐ ┌───▼───┐ ┌───────▼───────┐
      ┌───────▼───────┐    │             │
      │   LanceDB     │    │             │
      │  (Default)    │    │             │
      └───────────────┘    │             │
      ┌───────▼───────┐ ┌───▼───┐ ┌───────▼───────┐
      │   ChromaDB    │ │Ollama │ │    OpenAI     │
      │ (Alternative) │ │(Local)│ │  (Embeddings) │
      └───────────────┘ └───────┘ └───────────────┘

MCP Integration¶

Berry includes a native MCP (Model Context Protocol) server for AI assistant integration.

Claude Code¶

Add to your Claude Code MCP configuration:

{
  "mcpServers": {
    "berry": {
      "type": "stdio",
      "command": "berry",
      "args": ["mcp"]
    }
  }
}

Available Tools¶

Memory Types¶

Berry supports three memory types to help organize information:

Embedding Providers¶

Berry supports multiple embedding providers for semantic search:

OpenAI (Cloud)¶

export EMBEDDING_PROVIDER=openai
export EMBEDDING_MODEL=text-embedding-3-small
export OPENAI_API_KEY=sk-your-key

Ollama (Local)¶

ollama pull nomic-embed-text

export EMBEDDING_PROVIDER=openai
export EMBEDDING_MODEL=nomic-embed-text
export EMBEDDING_BASE_URL=http://localhost:11434/v1

Documentation¶

Requirements¶

• Embedding Provider: OpenAI API or local Ollama
• Rust 1.75+: Only required for building from source
• ChromaDB (optional): Only if using ChromaDB instead of the default LanceDB

Status¶

Beta: Berry is functional and safe to use, but breaking changes may occur before v1.0.

License¶

Berry is open source software. See the repository for license details.

Links¶

• GitHub Repository
• ChromaDB
• Ollama