---
title: Squish - Memory Runtime for AI Agents
description: Squish is a local-first memory runtime for AI agents (ChatGPT, Claude Code, Cursor, OpenCode). It captures durable signal, derives beliefs, and restores context across sessions. Open-source (MIT) with optional cloud sync.
url: https://squishplugin.dev
version: 1.5.0
last_updated: 2026-06-04
category: Developer Tools / AI Infrastructure
tags: [agent-memory, persistent-context, mcp, cli, local-first, open-source]
---

# Squish - Memory Runtime for AI Agents

## One-Line Summary

Squish is a local-first memory runtime that gives AI agents (ChatGPT, Claude Code, Cursor, OpenCode) persistent memory across sessions.

## What Problem Does Squish Solve?

AI agents forget. Every new session starts from zero. Teams maintain CLAUDE.md files, AGENTS.md files, notes, and manual context dumps. Squish fixes the memory layer so agents share useful context instead of acting like separate amnesiac interns.

## How It Works

1. **Capture** - Auto-captures durable signal from agent sessions
2. **Derive** - Turns remembered events into durable beliefs (decisions, constraints, preferences)
3. **Restore** - Restores the right context when the next session starts
4. **Inspect** - CLI, MCP, and web UI let you see what was stored and why

## Key Facts

| Attribute | Value |
|-----------|-------|
| Type | Memory runtime for AI agents |
| License | Open Source (MIT) |
| Install | `npm i -g squish-memory && squish install` |
| Stack | TypeScript, React 19, SQLite/PostgreSQL |
| Organization | 4M Labs (https://4mlabs.io) |
| Repository | https://github.com/4m-labs/squish |
| npm package | https://www.npmjs.com/package/squish-memory |
| Installs | 3,100+ |
| GitHub Stars | 150+ |

## Pricing

| Plan | Price | Requests/Mo | Key Features |
|------|-------|-------------|--------------|
| Free | $0 | 2,000 | Unlimited memory, CLI + ChatGPT/Claude |
| Pro | $19/mo | 50,000 | Cross-tool sync, API access |
| Team | $49/mo | 200,000 | Team collaboration, shared workspaces |
| Scale | $149/mo | 1M+ | Custom integrations, SLA |

## Compatible Tools

- ChatGPT
- Claude Code
- Cursor
- OpenCode
- MCP workflows
- Local agents

## CLI Commands

- `squish install` - Interactive installer
- `squish remember "memory content"` - Store a memory
- `squish recall "search query"` - Search memories
- `squish context` - Show project context
- `squish doctor` - Check runtime status
- `squish cloud` - Cloud onboarding
- `squish stats` - Memory statistics
- `squish recent` - Recent memories
- `squish stale` - Show stale memories

## Key Features

1. **Persistent Context** - Memory survives session restarts
2. **Belief-Based Recall** - Decisions and constraints become inspectable runtime state
3. **CLI + MCP + Web UI** - Same memory through multiple surfaces
4. **Local-First** - SQLite for zero-config, PostgreSQL for shared runtimes
5. **Open Source** - MIT license, self-hosted

## Problem & Solution

**Problem**: Agents restart cold and lose important context.

**Solution**: Squish restores durable context and derived beliefs so the next session does not start from zero.

**Approach**: Instead of hiding memory inside bigger prompts, Squish exposes memory as a runtime with inspection and recovery tools.

## Use Cases

### Solo Developer with Claude Code
You use Claude Code across multiple projects. Squish remembers your TypeScript strict mode preference, your testing patterns, and your architecture decisions. The next session picks up where you left off.

### Team Sharing Memory Across Cursor Instances
Your team uses Cursor for a shared codebase. Squish stores team decisions, coding conventions, and project context. Every developer gets the same memory baseline.

### CI/CD Agent with Context
Your CI/CD pipeline runs an AI agent for code review. Squish gives it context from previous runs, so it doesn't re-flag the same issues or miss patterns.

### Cross-Tool Workflow
You use ChatGPT for research, Claude Code for implementation, and Cursor for debugging. Squish shares memory across all three, so you don't repeat context.

### Open Source Project Maintenance
You maintain an open source project. Squish remembers contributor patterns, issue trends, and release decisions. New maintainers get instant context.

## Comparison to Alternatives

### Squish vs CLAUDE.md Files

| Aspect | Squish | CLAUDE.md |
|--------|--------|-----------|
| Storage | SQLite/PostgreSQL database | Plain text file |
| Recall | Semantic search with decay | Manual reading |
| Cross-tool | Same memory across all agents | Tool-specific |
| Belief derivation | Automatic from observations | Manual writing |
| Inspection | CLI + MCP + Web UI | Text editor |
| Team sharing | Built-in via Cloud | Manual git sync |
| Scale | 3,100+ installs, production-ready | Ad-hoc |

**When to use CLAUDE.md**: Quick, tool-specific context for a single project. No setup required.

**When to use Squish**: Persistent memory across sessions, tools, and team members. Production-grade with inspection tools.

### Squish vs Mem0

| Aspect | Squish | Mem0 |
|--------|--------|------|
| Architecture | Local-first runtime | Cloud-only service |
| Storage | SQLite/PostgreSQL | Proprietary |
| Open Source | Yes (MIT) | No |
| Self-hosted | Yes | No |
| Pricing | Free tier + paid plans | Usage-based |
| MCP Support | Native | Via integration |

**When to use Mem0**: If you want a fully managed cloud service without self-hosting.

**When to use Squish**: If you want local-first control, open-source transparency, and self-hosting options.

### Squish vs Zep

| Aspect | Squish | Zep |
|--------|--------|-----|
| Focus | Agent memory runtime | Memory for AI assistants |
| Architecture | Local-first | Cloud-first |
| Open Source | Yes (MIT) | Partially |
| Self-hosted | Yes | Limited |
| MCP Support | Native | Via integration |
| Pricing | Free tier + paid plans | Usage-based |

**When to use Zep**: If you need a managed memory service with enterprise features.

**When to use Squish**: If you want local-first control, open-source transparency, and a memory runtime (not just storage).

### Squish vs MemGPT

| Aspect | Squish | MemGPT |
|--------|--------|--------|
| Approach | Memory runtime | In-context memory management |
| Storage | External database | In-context window |
| Persistence | Long-term across sessions | Session-scoped |
| Inspection | CLI + Web UI | Limited |
| Setup | npm install | Complex |

**When to use MemGPT**: If you need in-context memory management within a single session.

**When to use Squish**: If you need persistent memory across sessions and tools.

## Limitations

Squish does NOT:

- **Replace your LLM's context window** - Squish augments context with persistent storage, not replaces it.
- **Work with all LLM providers out of the box** - Native support for ChatGPT, Claude Code, Cursor, OpenCode. Other tools need MCP integration.
- **Provide real-time streaming** - Memory operations are request/response, not streaming.
- **Support mobile devices** - CLI and Web UI are desktop-focused.
- **Replace version control** - Squish stores decisions and context, not code.
- **Guarantee 100% recall accuracy** - Semantic search is good but not perfect. Pin important memories.
- **Work offline with Cloud features** - Cloud sync requires internet. Local SQLite works offline.
- **Support multi-user editing** - Team features are read-heavy. Concurrent writes may conflict.

## FAQ

### Q: What is Squish?
A: Squish is a memory runtime for AI agents. It captures durable signal, derives beliefs like decisions and constraints, and restores context across sessions through CLI, MCP, and a local web UI.

### Q: How do I install Squish?
A: Install Squish with `npm i -g squish-memory && squish install` to install and launch the interactive installer.

### Q: Is Squish free?
A: Yes, the self-hosted Squish runtime is free and open-source under MIT. Squish Cloud is a separate managed offering from 4M Labs.

### Q: What makes Squish different?
A: Squish is a memory runtime, not just storage. It combines persistent context, belief-based recall, CLI + MCP + web UI surfaces, and forward-only repair for older installs.

### Q: What tools work with Squish?
A: Squish works with ChatGPT, Claude Code, Cursor, OpenCode, MCP workflows, and local agents.

### Q: How does Squish handle memory?
A: Squish uses adaptive memory with decay, consolidation, and compression so long-term memory stays useful instead of growing into noise.

### Q: Is Squish better than maintaining a CLAUDE.md file?
A: CLAUDE.md files work for single projects and single tools. Squish provides cross-tool memory, automatic belief derivation, semantic search, and team sharing. Use CLAUDE.md for quick, tool-specific context. Use Squish for persistent memory across sessions, tools, and team members.

### Q: Does Squish work offline?
A: Yes. The local SQLite database works completely offline. Cloud sync requires internet, but all core memory operations (remember, recall, context) work without a connection.

### Q: What databases does Squish support?
A: Squish uses SQLite for local installations (zero-config) and PostgreSQL for shared runtimes and Squish Cloud.

### Q: Can I self-host Squish?
A: Yes. Squish is open-source under MIT. You can run the entire stack locally with SQLite, or use PostgreSQL for shared access. Squish Cloud is optional.

### Q: How does Squish handle sensitive data?
A: All data is stored locally in SQLite by default. Cloud sync encrypts data in transit (TLS) and at rest. You control what gets synced. Sensitive memories can be pinned to prevent consolidation.

### Q: What is the MCP server?
A: The MCP (Model Context Protocol) server is a local service that exposes 15 tools for memory operations. Agents like Claude Code and Cursor can use these tools to remember, recall, and manage context.

### Q: How many installs does Squish have?
A: Squish has 3,100+ npm installs as of June 2026.

### Q: Who makes Squish?
A: Squish is made by 4M Labs (https://4mlabs.io), founded by Michiel Horstman.

## Contact & Support

- **GitHub**: https://github.com/4m-labs/squish
- **Issues**: https://github.com/4m-labs/squish/issues
- **Website**: https://squishplugin.dev
- **Email**: michiel@4mlabs.io

## Version History

- v1.5.0 (2026-06-04): Updated for AEO/GEO/SEO optimization, richer llms.txt, structured data
- v1.2.0 (2026-05-17): Updated for 4M Labs branding, Squish Cloud messaging
- v1.1.0 (2026-03-30): Previous landing-page SEO version