Persome
Local-first macOS personal model that gives MCP agents evidence-linked memory of you.
Open source
Persome is an open-source macOS runtime that captures focused cross-app activity after permission, builds an inspectable Point/Line/Face/Volume/Root model of one person, and exposes durable Markdown memory plus provenance tools to Claude Code, Codex, and other trusted MCP clients—private by default under ~/.persome.
Official resources
Selection advice
Choose Persome when agents should continue your work from a private macOS personal model with provenance. Choose Mem0 when memory belongs inside an application you ship, or EverOS when you want Markdown-native memory without OS activity capture.
Quick comparison
Persome builds a local personal model from macOS activity and serves it over MCP. Mem0 embeds memory inside application code, while EverOS keeps portable Markdown memory across coding assistants.
| Persome | Mem0 | EverOS | |
|---|---|---|---|
| Best for | Giving coding agents a private, evidence-linked model of one macOS user | Embedding user memory in a product you ship | Portable Markdown-native memory across local agent workflows |
| Storage model | Local ~/.persome captures plus Markdown memory and a layered personal model | SDK-managed memory with self-hosted and managed options | Markdown source of truth with SQLite and LanceDB indexes |
| Integration style | macOS runtime, CLI, and official MCP registry stdio/HTTP servers | Python and TypeScript APIs inside application code | Local Python runtime, HTTP API, and MCP clients |
| Tradeoff | Strong personal-context capture on macOS; not a cross-platform app memory SDK | Application-friendly API with less OS-level activity modeling | File-native portability without macOS Accessibility capture |
Best for
- local-first personal memory
- macOS coding-agent context
- evidence-linked MCP recall
- inspectable personal models
Not ideal for
- Linux or Windows-first agent deployments
- teams that need a managed multi-tenant memory SaaS
- products that only need short session caches
Core concepts
Point/Line/Face/Volume/Root modelHUMAN.md personal modelMCP memory toolsevidence receiptslocal-first privacy
Minimal implementation shape
Install with `uv tool install personal-model`, run `persome onboard`, then connect an MCP client with `persome mcp` (or `{"command":"persome","args":["mcp"]}`) so agents can call `search`, `list_memories`, and `get_model_snapshot`.