ai-memory

ai-memory keeps project context for developers who switch coding assistants or machines. It preserves unfinished work, failed approaches and open questions so the next session can pick up the task.

• Cross-agent handoffs: Claude Code, Codex, Cursor, Gemini CLI and OpenCode can share project memory and pass unfinished tasks between sessions.
• Editable Markdown wiki: Project knowledge lives in ordinary Markdown files backed by git, which you can edit directly or read in Obsidian; the search index can be rebuilt from those files.
• Automatic capture and retrieval: It records prompts, tool calls and session boundaries, sanitizes observations before storage, and compiles them into readable pages that agents can retrieve through session briefs and search by text, entities and links.
• Shared team knowledge: A shared server gives teammates access to project knowledge while keeping personal handoffs separate, with user authentication, individual attribution and an audit log.
• Cross-project preferences: A profile carries working preferences into new repositories while deferring to repository rules; profiles are private and opt-in on shared servers.

The MIT-licensed open source server runs on your laptop or a self-hosted machine. It supports Linux, macOS and Windows through WSL2. Capture, search and handoffs don't require LLM calls or an API key by default; LLM-written pages and vector search are optional. Repository capture rules can exclude sensitive paths or restrict capture to an allowlist. A read-only browser view lets you inspect the wiki outside your coding assistant.