aerovato/operator-memory

What it solves

Coding agents typically lose all context once a session ends, forcing them to waste tokens re-exploring codebases and re-learning architectural decisions in every new conversation. Operator Memory provides a durable, self-improving "brain" that allows agents to document and recall specifications, decisions, and lessons across sessions.

How it works

Instead of using a vector database or RAG, the system uses a workspace of Markdown documents stored in three locations: a private project folder (.operator/), a shared project folder (.operator-shared/), and a global user folder (~/.operator/user/). The agent follows a continuous loop: it consults these documents to gain context, performs development work, and then updates the Markdown files to record new truths, specs, or research.

Who it’s for

Developers using AI coding agents (such as Claude Code, Codex, OpenCode, Pi, or DeepSeek) who want their agents to maintain long-term project memory and shared team knowledge without managing complex infrastructure.

Highlights

• Automatic Documentation: The agent automatically records specs, research, and decisions as it works.
• Transparent Storage: Memory is stored as human-readable Markdown files that can be edited or deleted manually.
• Git Integration: Shared knowledge can be tracked via Git to synchronize memory across a team.
• Zero Infrastructure: Operates without the need for embeddings, vector databases, or background agents.