Gomaa
Gomaa is a local-first, hierarchical memory engine for autonomous AI agents that stores memories as human-readable Markdown notes in an Obsidian vault, backed by PostgreSQL with pgvector or zero-config SQLite, so a person can audit and edit what an agent has learned.
Gomaa targets three problems the project identifies in typical agent memory systems: memories that disappear into an opaque vector store a human cannot inspect or correct, unbounded accumulation of stale context that pollutes the prompt window, and different kinds of memory colliding with each other. Every memory is written as a Markdown note with wiki-links and frontmatter inside an Obsidian vault, classified into five cognitive layers such as episodic and procedural memory, and a temporal decay function fades inactive memories over time while pinned notes stay permanent.
A two-level scope of "wing" and "room" isolates context by domain and topic, and a multi-agent fleet can share sanitized global memories while keeping private databases separate. Retrieval combines hybrid reciprocal rank fusion search with a built-in dashboard for browsing the resulting knowledge graph.
It works with Hermes, OpenClaw, Claude Desktop, Cursor, Windsurf, CrewAI and LangChain, is written in Python, and is released under the Apache-2.0 license.