Skip to content
Kitbay
Submit
MCP server

agent-memory

Runs locallyMITUpdated on Nov 10, 20264,311GitHub starsCan change dataNo API key needed

agent-memory from Tigerless Labs is a long-term memory store for coding agents and other assistants.

What it does

agent-memory from Tigerless Labs is a long-term memory store for coding agents and other assistants. Memories are Markdown files in one folder, with a SQLite index beside them that can be rebuilt at any time; recall returns ranked file paths with one-line abstracts, and the agent opens only what it needs. A command-line tool records, corrects, merges and deletes memories, hooks capture them when a conversation ends, and a scheduled pass consolidates and prunes the store. An optional stdio MCP server offers the core calls. It installs from a Git checkout with uv and Python 3.12, works with Claude Code, Codex CLI and Muse Code, and needs no API key.

Capabilities

  • Recall ranked memories by query and read them at outline, abstract or full depth
  • Record, correct, supersede, merge and delete memories
  • Capture memories automatically at conversation boundaries through hooks
  • Consolidate and forget memories in a scheduled background pass
  • Share one memory store between several agent hosts