MemoryRouter

One memory for every AI you use. MemoryRouter lets people carry approved context from ChatGPT to Claude and other AI tools, while giving product teams private, persistent memory for every user through APIs and MCP.

Homepage · Start free · Documentation · Pricing · Security · Releases

Choose your path

I want to…	Start here
Move useful context from ChatGPT to Claude	Transfer your memories
Connect Claude, ChatGPT, Codex, OpenClaw, or another MCP client	MCP and connector guide
Give OpenClaw automatic persistent memory	OpenClaw setup
Add user-scoped memory to an AI product	Developer quickstart
Install the MemoryRouter CLI	Latest CLI release

Connect AI tools with MCP

MemoryRouter exposes a remote MCP server at:

https://mcp.memoryrouter.ai/mcp

Connect supported hosts through OAuth, choose a vault and permissions, then verify access with the host's visible MemoryRouter tools. Hosted connectors are model-directed: connecting does not automatically import old conversations or guarantee a memory call on every turn.

See the MCP overview for exact setup and host-specific behavior.

OpenClaw plugin

Install the current audited npm package:

openclaw plugins install npm:mr-memory@3.7.8
openclaw mr <your-memory-key>
openclaw mr status

The plugin provides relay-based recall and capture while inference and provider credentials stay inside OpenClaw. Historical workspace and session upload is a separate, explicit operation:

openclaw mr upload

Read the OpenClaw documentation before enabling it, especially if you use multiple agents or custom paths.

API quickstart

MemoryRouter supports two production patterns:

• Proxy mode: send the model request through MemoryRouter; it retrieves memory, calls the provider, and stores the completed exchange.
• Local inference mode: keep the provider call in your app and use /v1/memory/prepare plus /v1/memory/ingest for retrieval and storage only.

One stable Memory Key identifies one private user vault. Here is the OpenAI-compatible proxy path:

curl https://api.memoryrouter.ai/v1/chat/completions \
  -H "Authorization: Bearer $MEMORYROUTER_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-5.5",
    "messages": [
      {"role": "user", "content": "Remember that I prefer concise answers."}
    ]
  }'

For per-user key provisioning, local inference, native Anthropic and Google endpoints, imports, and deletion, use the developer quickstart and API reference.

CLI releases

The Releases page publishes MemoryRouter CLI binaries for macOS and Linux. Install the current release with:

curl -fsSL https://memoryrouter.ai/install.sh | sh
memoryrouter auth <your-memory-key>
memoryrouter status

The patches/ directory contains legacy OpenClaw plugin-API patch helpers for historical builds. Current OpenClaw users should follow the published npm installation path above instead of applying a legacy patch.

Pricing

MemoryRouter pricing and allowances vary by product path and can change. Use the live pricing page and the relevant integration guide rather than older README quota claims.

Support and trust

• Product and account help: hello@memoryrouter.ai
• Security policy: SECURITY.md
• Support policy: SUPPORT.md
• Contribution guide: CONTRIBUTING.md
• Privacy: memoryrouter.ai/privacy
• Terms: memoryrouter.ai/terms

Never post a live Memory Key, provider credential, or private memory content in a GitHub issue.

License

The public files in this repository are available under the MIT License.