Skip to main content

​
The Problem: Context Loss Between AI Sessions

Every new AI coding session starts from scratch. Your assistant has no memory of:
  • What you worked on yesterday
  • Bugs you’ve already fixed
  • Project conventions you’ve established
  • Code architecture decisions you’ve made
You waste valuable time re-explaining context that should already be known.

​
The Solution: LongMem

LongMem is a persistent memory daemon that runs locally on your machine. It records your coding activity and stores it in a private SQLite database, giving your AI assistant long-term memory across sessions.
Never repeat yourself to your AI again. LongMem keeps a private, local memory of your coding work so your assistant always starts with context.

​
Key Features

Local-First

All data stays on your machine by default. No cloud uploads unless you explicitly enable compression.

Automatic Context

Your AI assistant automatically recalls past sessions without manual prompting.

Smart Privacy

Built-in secret redaction protects sensitive data like API keys and credentials.

Zero Manual Work

No manual note-taking. LongMem hooks into your AI assistant automatically.

​
What LongMem Stores

Stored locally in ~/.longmem/memory.db:
  • User prompts to the AI
  • Commands executed by the AI
  • Tool outputs (after redaction)
  • File references and changes
Never uploaded unless you explicitly enable compression with a cloud provider.

​
How It Works

LongMem integrates seamlessly with Claude Code and OpenCode through hooks: 
Claude Code / OpenCode
        |
      hooks
        v
   +------------+
   |  longmemd  |  (local daemon)
   +------------+
        |
     SQLite DB
        v
  ~/.longmem/memory.db
  1. Your AI assistant runs commands - LongMem hooks capture the activity
  2. Events are sent to the daemon - Running locally on port 38741
  3. Data is stored in SQLite - Encrypted at rest, searchable, exportable
  4. AI queries memory via MCP - Automatic context retrieval in future sessions

​
Privacy Modes

LongMem offers three privacy levels:
ModeBehavior
safeRedacts common secrets + blocks sensitive files (recommended)
flexibleSafe mode + custom regex patterns for domain-specific secrets
noneNo redaction (for air-gapped or fully trusted environments)
Even in safe mode, review your ~/.longmem/memory.db periodically to ensure no sensitive data was captured.

​
Optional Compression

Compression creates short summaries of past sessions using an LLM, improving recall and search relevance.
  • Without compression: LongMem works fully, but may return raw verbose logs
  • With compression: Better semantic search, requires API key (OpenRouter, OpenAI, Anthropic, or local Ollama)
Compression is optional and disabled by default.

​
Next Steps

Installation

Install LongMem on macOS or Linux in 2 minutes

Quickstart

Get LongMem running in 5 minutes

Build docs developers (and LLMs) love

Get started for freeTalk to us