tigerless-labs/agent-memory
agent-memory keeps agent recall in plain Markdown files, with a rebuildable local index beside them.
The project positions files as the source of truth and SQLite/FTS/vector indexes as disposable cache. Recall returns ranked paths and anchors, then lets the agent read abstracts, outlines, or full files as needed. Writes are captured at conversation boundaries, with later consolidation that can update memories but only proposes deletions for confirmation. The repo says it needs no API key and can be shared across Claude Code, Codex CLI, MCP-capable agents, or any host that can run shell commands. GitHub · LLM repos' note.
The project positions files as the source of truth and SQLite/FTS/vector indexes as disposable cache. Recall returns ranked paths and anchors, then lets the agent read abstracts, outlines, or full files as needed. Writes are captured at conversation boundaries, with later consolidation that can update memories but only proposes deletions for confirmation. The repo says it needs no API key and can be shared across Claude Code, Codex CLI, MCP-capable agents, or any host that can run shell commands. GitHub · LLM repos' note.
score 4