← Back to Discover

agmem

Command-line tool that indexes a code repository into a local, searchable store so AI coding agents can pull task-relevant, source-linked context instead of re-searching every session.

Command-line toolsCoding agents
Visit website ↗

What it does

agmem builds a persistent project memory for coding agents. It walks the repository, respecting .gitignore, and extracts Terraform resources, Python and Go code, YAML and Helm values and Markdown sections into a local JSONL store. Asking for context on a task returns matching chunks grouped as constraints, facts and patterns, each linked to its source file and commit. Retrieval is keyword-based (BM25) with an optional local embeddings add-on, so it makes no LLM calls and needs no server. Entries carry source hashes so you can tell when memory has gone stale, and rules you record by hand rank higher in results. It works with any agent that can run a shell command, and the README describes a Claude Code plugin and git hooks that keep the store current. The project describes itself as alpha, with APIs that may change before 1.0. It is tested on macOS and Linux, and Windows support is best effort.

Capabilities

  • Indexes Terraform, Python, Go, YAML/Helm and Markdown
  • Returns constraints, facts and patterns with source references
  • BM25 search with optional local embeddings
  • Detects stale entries through source hashes
  • Hand-written rules that rank higher in results
  • Claude Code plugin and git hooks to keep memory in sync

When to use it

  • Your coding agent keeps re-searching the same repository each session
  • You want project context stored locally in plain, inspectable files
  • You want team conventions to reach Claude Code, Codex, Cursor or Aider through a shell command

Alternatives to explore

  • mem0
  • Letta
  • claude-mem