memento by Ddong8 is a self-hosted memory system designed to act as a shared brain for AI coding tools, capturing conversations and plans across agents. The app automatically collects and synchronizes chats, builds a knowledge graph, and exposes hybrid search so agents can recall long-term context. Key capabilities include BGE-M3 vector search, jieba full-text indexing, daily AI summaries, and local sanitization. It targets developers and power users who need persistent, private context across tools.

What tasks can you actually use it for?

memento functions as a persistent context store that developers use to recall prior decisions, search past conversations, and generate activity summaries. The tool ingests assistant chats from Claude Code, Cursor, Codex, Antigravity, and Obsidian and constructs a knowledge graph of entities, relations, and observations. That graph plus hybrid retrieval supports tasks such as context injection for prompts, project timeline reviews, and automated daily reports that summarize coding progress.

How reliable are the search and recall results?

The search pipeline combines BGE-M3 vector embeddings for semantic matches with a jieba-based full-text index for keyword lookups, so results blend semantic and lexical retrieval. The knowledge graph extracts entities and relations from chats, which improves targeted recall compared with raw retrieval alone. Retrieval quality depends on the clarity of source conversations and the extraction step; noisy or poorly segmented chats reduce precision in graph edges and vectors.

What inputs and infrastructure does it require?

Deployment requires a Model Context Protocol (MCP) host and a self-hosted backend orchestrated via Docker Compose, using Postgres, Redis, and MinIO. A local collector synchronizes across Mac, Linux, and Windows clients. Integration is MCP-ready, so the tool connects to MCP-compatible clients for in-place memory queries. Multi-tenant support, role-based access control, and audit logs fit team environments but add operational setup tasks.

How does it handle privacy, sanitization, and team access?

memento keeps data on your infrastructure and implements local sanitization that removes secrets before storage, which supports privacy-conscious workflows. The system offers multi-tenant tenancy with role-based controls and audit logging for team governance. Because the project is self-hosted, organizations retain file custody and can review logs and sanitization rules, aligning the tool with environments that require explicit data control rather than external hosting.

A practical match for infrastructure-savvy developers who need long-term context

memento is a practical choice for developers and power users who require persistent, searchable memory across multiple AI assistants and who can operate self-hosted services. The project is well-regarded in the open-source MCP community, which suggests active integration with MCP clients. Expect a learning curve for deployment and ongoing maintenance; this is best for teams that accept operational responsibility in exchange for tighter data control.