memento

Memento is a shared memory system for multiple AI agents that stores durable knowledge outside individual conversations. It addresses the need for persistent information about people, projects, machines, services, decisions and their relationships to remain available across agents while keeping transient data such as chats, reminders, credentials and machine-specific state with the owning agent.

Knowledge is stored as Markdown files with stable IDs, metadata and links. Git manages those files and their version history. A control.sqlite database holds proposals, idempotency records, write journals, leases and scheduler state. Separate full-text search, graph and optional vector indexes are derived from the Markdown content and can be rebuilt from source. A fine-tuned local Needle model classifies natural-language requests into a small set of read operations to reduce ambiguity. The system supports MCP access so any authenticated client can search, read and propose changes. Curators review all writes before they are committed to the shared repository.

Visual tools include a memory graph overview that displays concepts, services, links, clusters and diagnostics, a memory inspector showing selected concept metadata, links, assets, proposals and embedding state, and a large-graph stress view capable of rendering a 2,000-node graph with level-of-detail and runtime diagnostics. The project was originally built for several piclaw instances but is designed for broader use by any compatible MCP client.

Memento is an open-source project hosted on GitHub under the rcarmo account. No pricing information is stated.