Overview

ai-memory is a revolutionary solution designed to provide AI coding agents with persistent, long-term memory. It elegantly solves the problem of context loss when sessions end, enabling seamless handoffs between different agent vendors and even different instances of the same agent. By compiling a shared, persistent wiki from sanitized lifecycle observations, ai-memory ensures that your agents remember their past work, saving you from repetitive explanations of architecture, failed approaches, and open questions. The wiki is stored as plain markdown in a Git repository, making it easily searchable, accessible with tools like Obsidian, and securely backed up.

Key Features

• Effortless Lifecycle Capture: Hooks automatically capture bounded, sanitized observations of prompts, tool lifecycles, and session boundaries, keeping your memory lightweight and focused.
• Seamless Managed Workstreams: Initiate and resume logical workstreams across different agents (e.g., Claude Code to Codex) with native per-harness session continuity and a portable visible-event ledger.
• Per-Repository Capture Exclusions: Granularly define which files or directories should be excluded from memory capture, ensuring privacy and relevance.
• Cross-Agent Handoffs: Gracefully pause tasks with one agent and resume them with another, picking up exactly where you left off without re-explaining context.
• Project Isolation by Construction: Each project is uniquely identified and isolated within the wiki, preventing collisions and simplifying management, even in complex monorepos or multi-client environments.
• Global Preferences Scope: Store and access standing user or team context (tech choices, code style, durable rules) that travels with you across projects.
• Entity-Assisted Recall: Recover pages using exact, prefix, or compound-word matches on specific entities declared in the page's frontmatter, enhancing searchability beyond simple keywords.
• Authority-Aware Recall: A sophisticated retrieval system combines FTS5, entity matching, graph-neighbor analysis, and optional vector search, with a bias towards authoritative pages like _rules/ and decisions/.
• Clear Routing Alongside Code-Intelligence Tools: Run ai-memory independently of other code-intelligence tools, using it for historical context while relying on live tools for structural code analysis.
• Karpathy-style LLM Wiki: Pages are compiled at session-end, not retrieved over raw logs, enabling time-travel capabilities through Git history and versioned markdown.
• Built-in Web Browser: Access a read-only HTML UI for browsing the wiki, searching, and viewing markdown content.
• Multi-Agent & Multi-Machine Ready: Supports a wide array of AI coding agents and can be deployed on local machines or a central server with secure token-based authentication.
• Thin-Client CLI: A comprehensive command-line interface for managing memory, including bootstrapping, checkpoints, restoration, auditing, and more, all communicating with the central server.
• LLM is Opt-In: Utilize powerful FTS5, entity, and graph-neighbor search capabilities even without an LLM provider. Integrate LLMs for advanced features like consolidated pages and automated improvements when desired.

Typical Use Cases

• "Quit Claude Code and continue the same work in Codex." Seamlessly transition your AI agent's work. Start a session with one agent, then pick up the exact same task with a different agent without missing a beat.
• "Pick the project instead of remembering where it lives." Effortlessly manage multiple projects. A simple ai-memory show command lists your projects, allowing you to select and launch into any of them, even starting new projects with automated setup.
• Context Persistence: Ensure all your AI coding sessions, regardless of the agent or duration, contribute to a rich, persistent knowledge base that agents can leverage for future tasks.
• Knowledge Sharing and Handoffs: Facilitate smooth handoffs between team members or between different AI agents by providing a clear, consolidated summary of the work done, including decisions, failed attempts, and next steps.