Long-term memory for coding agents that stores memories in a local SQLite database and shares them across Claude Code, Codex, Cursor and other MCP clients.
CodeAbra/iai-personal-memory-engine
A local MCP memory server that records every AI coding session word-for-word and feeds relevant past context back to agents like Claude Code and Cursor.
- Stars
- 917
- Forks
- 111
- Open issues
- 16
- Last commit
- 4 days ago
What is it?
iai-memory is a local server that speaks the Model Context Protocol and gives Claude and other MCP-compatible agents a long-term memory. It captures every turn of each session verbatim, organizes those captures over time into semantic summaries and behavioural parameters, and returns a small slice of relevant memory at the start of new conversations and turns, so users don't need to ask the agent to remember things.
Records are encrypted at rest with AES-256-GCM and stored under ~/.iai-mcp/. The embeddings and the store run on the local machine, and there is no account or hosted service. Some network calls remain: normal model calls from the CLI, an optional nightly consolidation step through the user's Claude subscription (can be disabled with an environment variable), a daily version check, and a one-time embedding model download.
The project includes a local dashboard (iai brain) and a CLI for searching, capturing, and uploading documents. Host support varies: Claude Code gets the fullest ambient integration, while Codex, Cursor, Hermes, and OpenClaw have different levels of hook support. macOS and Linux are supported, and Windows is in beta.
Key features
- 01Captures every session turn verbatim and stores it encrypted locally.
- 02Recalls relevant memory at session start and on later turns via MCP tools.
- 03Combines semantic similarity, graph links, recency, and lexical matching for retrieval.
- 04Tracks contradictions by closing old records' validity and linking the replacement.
- 05Ships a local dashboard and CLI for search, pinning, and document upload.
Use cases
- Keeping context across many Claude Code sessions without repeating project background.
- Giving Cursor or Codex agents access to earlier decisions and conversations on your own machine.
- Searching past coding conversations by exact wording.
- Ingesting local documents and code so an agent can recall them later.
Quick start
- 1
Install the Python package from PyPI
python3.12 -m pip install -U iai-pme - 2
In Claude Code, add the plugin marketplace
/plugin marketplace add CodeAbra/iai-personal-memory-engine - 3
In Claude Code, install the iai-memory plugin, then restart the session and verify with iai-mcp doctor
/plugin install iai-memory@iai-pme
Things to know
- Windows support is beta and ambient behaviour varies by host.
- The nightly REM step is on by default and may call claude -p, so it is disabled only with IAI_MCP_REM_DISABLED=1.
- The project is solo-maintained and has no enterprise SLA.
- Recall is usually mediocre during roughly the first ten sessions.
In collections
Topics
FAQ
- What is iai-personal-memory-engine?
- A local MCP memory server that records every AI coding session word-for-word and feeds relevant past context back to agents like Claude Code and Cursor. iai-memory is a local server that speaks the Model Context Protocol and gives Claude and other MCP-compatible agents a long-term memory. It captures every turn of each session verbatim, organizes those captures over time into semantic summaries and behavioural parameters, and returns a small slice of relevant memory at the start of new conversations and turns, so users don't need to ask the agent to remember things.
- Is iai-personal-memory-engine free to use? What license does it have?
- iai-personal-memory-engine is open source under the MIT license. Check the license terms before using it in commercial products.
- Is iai-personal-memory-engine still maintained?
- Actively maintained. The last commit to iai-personal-memory-engine was on Nov 2, 2026.
- What can I use iai-personal-memory-engine for?
- Keeping context across many Claude Code sessions without repeating project background. Giving Cursor or Codex agents access to earlier decisions and conversations on your own machine. Searching past coding conversations by exact wording. Ingesting local documents and code so an agent can recall them later.
- How do I install iai-personal-memory-engine?
- Install the Python package from PyPI: python3.12 -m pip install -U iai-pme In Claude Code, add the plugin marketplace: /plugin marketplace add CodeAbra/iai-personal-memory-engine In Claude Code, install the iai-memory plugin, then restart the session and verify with iai-mcp doctor: /plugin install iai-memory@iai-pme
- Where can I get iai-personal-memory-engine?
- The source code and installation instructions are on GitHub: https://github.com/CodeAbra/iai-personal-memory-engine
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
Local-first, governed memory layer that gives AI agents like Claude Code, Cursor and Codex persistent recall that admits when it doesn't know.
A local Rust tool that gives Claude Code persistent memory of past conversations via semantic search, exposed as an MCP server.
Local-first persistent memory for AI coding agents, storing decisions and lessons in SQLite and serving them over MCP to Claude Code, Cursor and other clients.
Local-first search over your Claude and Codex session history, with a Rust search core, JSON CLI, and desktop app.
Local-first persistent memory for AI agents, stored as markdown files with a knowledge graph, retrieval, and an MCP server.