Rememble

Local-first memory server with hybrid search, knowledge graph, and RAG context assembly.

SQLite + sqlite-vec + FTS5 backend. Works with any MCP client.

Features

• Hybrid search: BM25 + vector KNN + temporal scoring (RRF fusion)
• Knowledge graph: entities, observations, relations
• Token-budgeted RAG context assembly
• Multiple embedding providers: Ollama, local (sentence-transformers), OpenAI-compat

Install

uv tool install rememble

Upgrade:

uv tool upgrade rememble

MCP Client Setup

Claude Code / Claude Desktop

{
  "mcpServers": {
    "rememble": {
      "command": "rememble"
    }
  }
}

Configuration

Config file: ~/.rememble/config.json (auto-created on first run).

All fields can be overridden via env vars with REMEMBLE_ prefix and __ as nested delimiter.

Embedding Providers

Ollama (default)

{ "embedding": { "provider": "ollama", "model": "nomic-embed-text", "dimensions": 768 } }

Local (sentence-transformers, no network)

uv sync --extra local

{ "embedding": { "provider": "local" } }

OpenAI-compat (OpenRouter, OpenAI, Cohere compat, etc.)

OpenRouter:

{
  "embedding": {
    "provider": "compat",
    "api_type": "openrouter",
    "api_endpoint": "https://openrouter.ai/api/v1",
    "model": "openai/text-embedding-3-small",
    "dimensions": 1536
  }
}

Cohere via OpenAI compat API:

{
  "embedding": {
    "provider": "compat",
    "api_type": "cohere",
    "api_endpoint": "https://api.cohere.com/compatibility/v1",
    "model": "embed-english-light-v3.0",
    "dimensions": 384
  }
}

Set REMEMBLE_EMBEDDING__API_KEY or OPENROUTER_API_KEY / OPENAI_API_KEY env var.

Development

make dev      # install deps
make test     # run tests
make check    # fmt + lint + test