Squish — Runtime de Memória para Agentes IA
michielhdoteth/squish
Memória persistente e pesquisável para agentes IA — local-first, sem APIs externas
Ver no GitHubMCP server local-first que fornece memória persistente, durável e pesquisável para agentes de IA entre sessões. Integra-se com Claude Code, Cursor, Codex e ferramentas compatíveis com MCP, eliminando a necessidade de re-injetar contexto manualmente a cada execução. Usa embeddings embutidos, grafo de conhecimento e recuperação híbrida sem exigir banco de dados externo ou chaves de API.
Instalação
npx convez-ai michielhdoteth/squish1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Persistir decisões arquiteturais e preferências entre sessões de coding agents
- Manter histórico pesquisável de execuções passadas de agentes
- Reduzir token waste evitando re-injeção manual de contexto repetido
- Sincronizar memória compartilhada entre múltiplos agentes e máquinas
Veredito

Squish resolve um problema real: agentes IA esquecendo contexto entre sessões. Solução local-first bem arquitetada com embeddings nativos e decay automático de memória stale. Reduz custo de tokens significativamente e funciona com múltiplos agentes sem lock-in.
README
A tradução para o português deste README ainda está na fila.
Squish — Memory Runtime for AI Agents
4.5k+ npm downloads. Local-first. Optional LLM support.
Star the repo if Squish saves you context.
Squish gives agents stable orientation, durable memory, and searchable session history across runs. Local-first. Optional LLM support. Works with Claude Code, Cursor, Codex, Copilot, Gemini CLI, and any MCP-compatible tool.
npm install -g squish-memory && squish install --all
Squish provides AI agent memory that persists between sessions, across agents, and across machines. It is a local-first MCP server with built-in embeddings, a knowledge graph, and hybrid retrieval — no external database or API key required. Use it for free locally, or enable Squish Cloud for cross-device sync.
Core Concepts
The Problem: Agents Forget Everything
Every AI coding agent starts from zero when a new session begins. The architecture decision from last week, the config you spent an hour debugging, the preference you mentioned yesterday — gone.
Built-in memory files like CLAUDE.md and .cursorrules help, but they have hard limits. They cap out around 200 lines, require manual curation, and do not work across agents. You end up copy-pasting the same context into every tool.
Squish gives you persistent memory for coding agents that scales without limits. No manual maintenance. No token waste. No agent lock-in.
Three Layers of Memory
Token Cost Comparison
Squish retrieves only the relevant memories for the current task. The average context injection is 50-200 tokens — a fraction of what you would paste manually.
Quick Start
Step 1: Install
npm install -g squish-memory && squish install --all
This installs the Squish CLI, MCP server, and plugin hooks for all detected agents.
Step 2: Work
Start your coding agent as usual. Squish runs in the background, auto-capturing decisions, constraints, preferences, and context.
squish remember "We chose PostgreSQL for team mode" --type decision
squish recall "project decisions"
Step 3: Search Past Sessions
After a few sessions, search your agent history:
squish sessions search "postgres migration"
squish sessions related --repo-path .
Step 4: Restart
Close your session and open a new one. Your agent picks up where you left off — all context is restored automatically.
squish context # See what your agent remembers
squish stats # Check memory health
Works locally free. Optional cloud sync available at squishplugin.dev.
Works with Every Agent
Squish works with any AI coding agent that supports MCP (Model Context Protocol) or HTTP connections. One memory server, shared across all of them.
Works with any agent that speaks MCP or HTTP. One server, memories shared across all of them.
MCP Server Configuration
Add Squish to any MCP-compatible client:
{
"mcpServers": {
"squish": {
"command": "squish-mcp",
"args": ["--http", "--port", "8767"],
"env": {
"SQUISH_DB_PATH": "./squish-data"
}
}
}
}
For cloud-connected agents:
{
"mcpServers": {
"squish-cloud": {
"type": "url",
"url": "https://api.squishplugin.dev/mcp",
"headers": {
"Authorization": "Bearer <your-token>"
}
}
}
}
Why Squish
Most memory tools need a second LLM for embeddings and retrieval. That means extra API costs, latency, and infrastructure you have to manage.
Squish uses local embeddings by default. Zero LLM dependency. 1-5ms latency. $0 runtime cost in local mode.
Bring your own LLM if you want — Squish supports external embeddings and reasoning, but nothing requires it.
Comparison
Features
Memory Intelligence
- Auto-captures decisions, constraints, and preferences as you work
- Restores relevant context when an agent restarts
- Handles contradictions and temporal facts with expiration
- Graph-boosted retrieval connects related memories across sessions
- Contradiction detection flags conflicting information
- Temporal reasoning tracks when facts were true vs. now
- Confidence scoring adjusts memory relevance over time
- Decay system automatically ages low-value memories
Session Search
- Search previous Claude Code, Codex, and OpenCode sessions
- Find related sessions by project path or file overlap
- Inspect past decisions, errors, and commands as evidence
- Separate from long-term memory — raw session history, not distilled facts
Interfaces
- CLI:
squish remember,recall,inspect,context,stats,search,sessions - MCP Server: 15 tools for any MCP client — recall, health, graph, recency, maintenance
- Web UI: Local dashboard at
localhost:37777for visualizing memories - Cloud Dashboard: Analytics and management at squishplugin.dev
Storage
- SQLite (local, default) or PostgreSQL (team mode)
- Hybrid retrieval: keyword + semantic similarity with RRF fusion
- AES-256-GCM encryption for sensitive memories
- Places routing: organize memories by project, feature, or context
- Full-text search with BM25 ranking
- Vector search with TF-IDF embeddings (768-dimensional)
Memory Pipeline
Squish uses a 4-stage pipeline to process memories:
- Capture — Filters noisy tool output, promotes what matters (decisions, constraints, preferences)
- Filter — Deduplicates, resolves contradictions, scores importance
- Store — Persists to SQLite/PostgreSQL with graph relationships and embeddings
- Retrieve — Hybrid search combines keyword, semantic, recency, and importance scoring
Architecture
Agent Action
|
v
[1. Capture] -----> Filter noisy output
| Promote decisions, constraints, preferences
v
[2. Store] -------> SQLite / PostgreSQL
| Embeddings (local TF-IDF)
| Knowledge graph edges
| Places routing
v
[3. Retrieve] ----> Keyword search (BM25)
| Semantic search (cosine similarity)
| Recency weighting
| RRF fusion scoring
v
[4. Context] -----> Inject relevant memories into agent context
50-200 tokens average
Auto-decay old/low-value memories
Three-Layer Memory Model
+------------------+ +------------------+ +------------------+
| RECALL | | SESSIONS | | REMEMBER |
| (durable) | | (evidence) | | (write) |
| | | | | |
| Decisions | | Past agent runs | | Store new facts |
| Preferences | | Searchable | | Auto-classify |
| Constraints | | Raw history | | Graph update |
| Beliefs | | Related repos | | Place routing |
+------------------+ +------------------+ +------------------+
| | |
v v v
squish recall squish sessions squish remember
squish_recall search/show/list squish_remember
Storage Layer
SQLite (default) PostgreSQL (team mode)
| |
v v
+------------------+ +------------------+
| memories | | memories |
| associations | | associations |
| embeddings | | embeddings |
| graph_edges | | graph_edges |
| places | | places |
| sessions | | sessions |
+------------------+ +------------------+
| AES-256-GCM | | AES-256-GCM |
| encryption | | encryption |
+------------------+ +------------------+
Squish Cloud
Persistent memory across ChatGPT, Claude Desktop, Claude Code, and local agents. One account, synchronized everywhere.
ChatGPT Claude Desktop Claude Code Local Agents
[OAuth 2.1] [OAuth 2.1] [Streamable HTTP] [MCP / CLI]
+-------------------+---------------+------------------+
|
Squish Cloud API
|
[PostgreSQL + Encrypted Storage]
|
Admin Dashboard & Analytics
Cloud features: OAuth 2.1 + PKCE login, cross-platform sync, team workspaces, admin dashboard, priority support.
Pricing
Sign up at squishplugin.dev — 30 seconds, no credit card needed.
Founder Pass is a launch-only offer. $99/year instead of $348/year (Pro monthly).
Installation Guides
- Claude Code — MCP server + plugin hooks for auto-capture
- OpenCode — MCP server + hooks for OpenCode agent
- OpenClaw — MCP server setup for OpenClaw
Quick install for all detected agents:
npm install -g squish-memory && squish install --all
Squish auto-detects which agents you have installed and configures hooks for each one.
Benchmarks
Squish is tested against real-world memory retrieval tasks and synthetic benchmarks.
Full benchmark details: docs/BENCHMARK.md
Documentation
FAQ
What is Squish?
Squish is a local-first memory runtime for AI coding agents. It gives your agents stable orientation, durable memory, and searchable session history across runs. Think of it as a brain that persists between sessions — your agents remember decisions, constraints, preferences, and context without you having to re-explain everything. In v1.6.0, Squish also searches past agent sessions as evidence, so agents can inspect prior work instead of starting from zero.
Does Squish require an API key?
No. Squish works locally by default with zero API keys. It uses local embeddings (TF-IDF) and SQLite storage. You can optionally configure an external LLM for enhanced reasoning, but it's not required. An API key is only needed if you want to use Squish Cloud for cross-device sync.
How does Squish compare to mem0 or agentmemory?
Squish is the only option that works locally with zero external dependencies. mem0 requires Qdrant (a vector database) and cloud API calls. agentmemory requires iii-engine. Squish uses SQLite and local embeddings by default. See the full comparison in the Why Squish section above.
Can I use Squish with multiple AI agents?
Yes. Squish works with any MCP-compatible agent. One memory server is shared across Claude Code, Cursor, Codex, Copilot, Gemini CLI, and any other agent that supports MCP. Memories are available to all connected agents.
Is my data private with Squish?
Yes. In local mode, all data stays on your machine in an encrypted SQLite database. Nothing is sent to any cloud service. AES-256-GCM encryption protects sensitive memories. In cloud mode, data is encrypted in transit and at rest.
What databases does Squish support?
Squish supports SQLite (default, local) and PostgreSQL (team mode). SQLite requires zero configuration. PostgreSQL is used for team workspaces and shared memory across multiple users.
What is the difference between recall and sessions?
squish recall searches your long-term memory — distilled facts, decisions, and preferences that Squish has captured and organized. squish sessions search searches raw past agent runs — the actual messages, commands, and file changes from previous Claude Code, Codex, or OpenCode sessions. Recall gives you what the system decided to remember. Sessions give you the evidence.
Contributing
See docs/CONTRIBUTING.md for guidelines on how to contribute to Squish.
License
MIT — see LICENSE for details.