memra-local

Local-first memory server for AI agents. Offline, private, fast.

memra-local gives your coding agent persistent memory that lives entirely on your machine — no account, no network, no data leaves the laptop. Works with Claude Code, Cursor, Zed, Droid, Hermes Agent, OpenClaw, and any MCP-compatible client.

When you're ready to sync across devices or share with a team, a single command pushes your local namespace to Memra Cloud. Same tools, same API, your choice.

Install

pip install memra-local
memra mcp          # start the MCP server

Requires Python 3.10+.

Wire it into your editor

Claude Code / Cursor

{
  "mcpServers": {
    "memra": {
      "command": "memra",
      "args": ["mcp"]
    }
  }
}

Zed

{
  "context_servers": {
    "memra": {
      "command": { "path": "memra", "args": ["mcp"] }
    }
  }
}

Droid (Factory.ai) / Hermes Agent / OpenClaw

See usememra.com/install for client-specific snippets.

What you get

• Flat-file memory in ~/.memra/ — plain YAML, inspectable, greppable, diff-able
• MCP server exposing memra_add, memra_recall, memra_get, memra_list, memra_supersede, memra_history, and more
• Local embeddings via fastembed (ONNX multilingual-e5-small, ~100 languages) — no OpenAI key, no PyTorch
• Sync to cloud optional: memra sync enable <namespace> --api-key memra_live_...

Commands

memra mcp          # MCP server over stdio
memra status       # store health: scope, memory count, disk usage, sync
memra hooks install  # optional — auto-capture decisions/patterns as you work
memra --help       # full CLI reference

Docs + source

• Install snippets and client configs: https://usememra.com/install
• Memra Cloud (hosted EU): https://usememra.com
• Source: https://github.com/usememra/memra-local

License

BUSL-1.1. Change Date 2030-04-17 — on that date the license auto-converts to Apache-2.0. Until then, personal and non-production use are unrestricted; commercial production use requires a separate license.