Overview

ai-memory provides long-term, persistent memory for AI coding agents, enabling seamless handoffs and context continuity across different vendors and sessions. Say goodbye to re-explaining architecture, failed approaches, or open questions when switching between agents like Claude Code, OpenAI Codex, or Command Code within the same project directory. It intelligently compiles sanitized lifecycle observations into a shared, persistent wiki (plain Markdown in a Git repo) that is easily searchable and manageable.

Key Features

• Effortless Context Continuity: Transition between AI coding sessions and agents without losing critical context. ai-memory ensures your next agent understands the project's history, prior decisions, and ongoing tasks.
• Zero-Friction Lifecycle Capture: Automatically and unobtrusively captures essential observations from agent interactions, including prompts, tool usage, and session boundaries. This data is sanitized and stored efficiently.
• Managed Workstreams: Opt-in to managed workflows that provide native per-harness session resumption, a portable visible-event ledger, and comprehensive ledger search for higher-fidelity cross-harness continuity.
• Per-Repository Capture Exclusions: Fine-tune what gets captured by defining ignore policies within your repository, preventing sensitive or irrelevant information from cluttering the memory.
• Cross-Agent Handoffs: Enables agents to pick up exactly where another left off, presenting relevant context before their initial prompt, crucial for complex or lengthy development tasks.
• Per-Project Isolation: Organizes memory data with stable UUIDs for projects and workspaces, ensuring clear separation even for complex monorepos or multiple client projects within a single checkout.
• Global Preferences: Store and recall standing user or team context, like coding standards or durable personal rules, in a dedicated "global" scope that's automatically integrated into project queries.
• Entity-Assisted Recall: Enhances recall by storing and matching specific entities (nouns) within page frontmatter, allowing for accurate retrieval even when wording differs.
• Authority-Aware Recall: Leverages multiple search mechanisms (FTS5, entity-match, graph, optional vector) and prioritizes specific types of pages (rules, decisions) while still making historical session data discoverable.
• Karpathy-Style LLM Wiki: Pages are compiled at session-end, not retrieved from raw logs, facilitating time-travel capabilities with Git versioning and built-in checkpoint management.
• Built-in Web UI: Access a read-only HTML interface for browsing projects, searching, rendering Markdown, and managing your memory.
• Multi-Agent & Multi-Machine Ready: Supports a wide range of popular AI coding agents and is designed for both local and networked deployments with robust authentication.
• Thin-Client CLI: Interact with the ai-memory server through a lightweight CLI for various management tasks without direct access to underlying storage.
• Opt-In LLM Integration: Choose to use LLM capabilities for advanced features like consolidated pages and auto-improvement, or operate in a zero-LLM mode with core FTS5 and rule-based summarization.

Typical Use Cases

• Seamless Agent Switching: "Quit Claude Code mid-task, start OpenAI Codex in the same directory hours later – and Codex will seamlessly pick up where you left off."
• Project Discovery and Resumption: Easily navigate and resume work on different projects by simply being in their directory. ai-memory show provides a clear list of discoverable projects and allows for the creation of new ones.
• Contextual Development: Utilize ai-memory alongside live code intelligence tools. Use memory for past decisions and rationale, while tools handle current code structure and dependencies, ensuring historical context is verified against the live checkout.
• Time-Traveling Code History: Revisit past states of your project's memory with ai-memory checkpoints and restore-page commands, leveraging Git's versioning for comprehensive history tracking.