Squish gives AI agents persistent memory that survives sessions, decays what goes stale, and stays on your machine. Free, open-source, MCP-native.
Squish remembers decisions, constraints, and preferences across sessions and restores them automatically when a new session starts. Semantic recall at 6.1ms, 39 ops/sec, all running locally.
Prompt files go stale and burn tokens. Vector databases retrieve raw text by similarity. Squish derives durable beliefs from observations, applies memory decay and consolidation, and keeps every memory inspectable and editable.
The core runtime is MIT-licensed and runs against a local SQLite database with zero cloud dependency. Optional Squish Cloud adds sync and team features starting at $9/month.
Install once and Claude Code, Cursor, Codex, OpenCode, and any MCP-compatible agent share the same persistent memory.
Because stateless agents waste your best work. Every session starts from zero, so you re-explain architecture decisions, preferences, and known bugs. Persistent memory lets an agent keep durable context across sessions and act on what it already knows.
Prompt files are manual, stale, and token-hungry; vector databases only retrieve raw embeddings by similarity. Squish is a memory runtime: it derives beliefs from observations, applies decay and consolidation, and keeps every stored memory inspectable and editable.
Yes. The local runtime is MIT-licensed with unlimited local memory and requests. Optional Squish Cloud adds sync and team features starting at $9/month.
Squish works with Claude Code, Cursor, Codex, OpenCode, Windsurf, and any MCP-compatible agent. Install once and every tool shares the same memory.