Shared memory for AI agents
Give every AI tool the same memory.
Agent Memory Hub stores project context, preferences, domain knowledge, and handoff state in a private Markdown/Git repository, then makes that memory available to your AI tools through MCP or OpenAPI.
Quick Start
- Create a private GitHub repository for memory, such as
agent-memory. - Deploy the Agent Memory Hub service to Render or another HTTPS host.
- Open
https://your-service/setupand copy the connection URL into your AI platform. - When starting a project session, ask your AI to use
session_capsule detail=compact.
Platform Connections
| Connection Type | Endpoint | Best For |
|---|---|---|
| Remote MCP | https://your-service/mcp |
Claude connectors, ChatGPT Apps or custom MCP, and other MCP-aware clients |
| OpenAPI Actions | https://your-service/openapi.json |
Custom GPT Actions and function-calling platforms |
| Local stdio MCP | python3 -m agent_memory_hub ... mcp |
Codex, Claude Code, and local desktop or CLI agents |
Safety Boundaries
- GitHub Pages hosts only this public introduction, never your private memory.
- Your memory API must use HTTPS and a bearer token.
- Never store API keys, private keys, recovery codes, payment details, or sensitive personal information.
- Memory writes, replacements, and retirements should require user confirmation; checkpoints only generate suggested changes.
Roadmap
This public page is intentionally minimal, focusing first on access and setup. Future versions can add one-click configuration, connection diagnostics, platform-specific configuration snippets, and visual memory management.