Infinite memory for Claude Code. Install once — every session remembers the last. Automatically.
Auto-recall at start. Auto-observe during work. Auto-save at end. Zero commands. Zero config. Zero friction.
You spend 20 minutes explaining your project architecture to Claude Code. Your folder structure, your naming conventions, the reason you picked one ORM over another.
Next session: blank slate. Claude Code does not remember any of it.
Other memory tools exist, but they require you to manually call commands every session — recall, observe, save. That is ironic — a memory system you have to remember to use.
SuperLocalMemory v3.3.6 solves this. Install once. Memory works automatically, across every session, forever.
Hooks auto-install. MCP auto-configures. Memory auto-activates. You do nothing.
This installs the engine, Python dependencies, Claude Code hooks, and creates the data directory. Open Claude Code — memory is active.
SessionStart, PostToolUse, Stop — all wired into settings.json
Claude Code discovers SLM tools automatically
Every new session loads your project context before your first message
Every Write and Edit is silently captured with rate-limiting
Closing a session saves a summary with git branch, commits, and modified files
npm update triggers silent hook upgrade — no manual intervention
If hooks don't auto-configure your IDE, add this to your MCP settings:
{
"mcpServers": {
"superlocalmemory": {
"command": "slm",
"args": ["mcp"],
"type": "stdio"
}
}
}Six hooks that make memory invisible. You never think about it — it just works.
SessionStart hook loads your project context, recent decisions, and learned patterns before your first message. Zero commands needed.
PostToolUse hook silently captures file changes with rate-limiting. Every Write and Edit is observed — your coding decisions are remembered.
Stop hook generates a session summary with git context — branch, commits, modified files. Next session picks up exactly where you left off.
After 20+ sessions, SLM optimizes retrieval for YOUR patterns. After 200+, a full ML model trains on your usage. Zero token cost.
Every hook wrapped with error absorption. Crashes are invisible. Hooks never block tools. Opt out anytime with one command.
Hooks install on npm install and upgrade silently on version change. Old users and new users get the same zero-friction experience.
SessionStart hook fires. Your project context, recent decisions, and learned patterns are injected before your first message. You see: "I recall you're working on the auth migration..."
PostToolUse hook silently observes every file edit. Architecture decisions, naming choices, and bug fixes are captured. No commands needed.
Stop hook saves a summary: git branch, 12 commits, 8 files modified. The auth migration context is locked in.
SessionStart fires again. Full context from the morning is there. Claude Code picks up exactly where you left off — no re-explanation needed.
Switch to a new repo. SLM recalls that project's context instead. Each project has its own memory profile.
Builds relationships between concepts, files, and decisions. Claude Code understands how parts of your codebase connect.
Local SQLite with 5-channel hybrid search. No network latency, no API rate limits. Instant memory retrieval.
Built on the Model Context Protocol. No custom integrations — Claude Code speaks MCP natively. 30+ tools available.
Your code context never leaves your machine. No cloud APIs, no telemetry, no third-party servers. Works air-gapped.
Best-in-class for any local-only system on the standard long-conversation memory benchmark. No cloud dependency.
Same memory works with Cursor, Windsurf, VS Code, ChatGPT Desktop, and more. Set up once, remember everywhere.
Folder structure, module boundaries, dependency choices, and the reasoning behind each one.
Naming patterns, formatting preferences, TypeScript strictness, import ordering, testing conventions.
Known issues, workarounds, edge cases investigated, bugs fixed and the root causes found.
What you worked on in each session — git branches, commits, modified files, and progress.
How you like explanations, your communication style, tools you prefer, approaches you've approved.
Trade-offs discussed over multiple sessions. Why you chose one library, rejected another, pivoted on a third.
Beyond auto-memory, SLM gives Claude Code a full memory toolkit.
session_init Load full context at session start
remember Store knowledge with importance scoring
recall 5-channel hybrid search across all memories
observe Silently capture content worth remembering
search Full-text search across memory graph
build_graph Build knowledge graph from memories
get_learned_patterns Retrieve patterns learned from your usage
consolidate_cognitive Merge related memories automatically
get_soft_prompts Context-aware prompts from your history
recall_trace Trace how a memory was retrieved
fetch Direct memory access by ID
get_behavioral_patterns Patterns detected from your workflow
The same memory database works across all MCP-compatible tools. Install once, remember everywhere.
Open source. AGPL v3. One command to install. Infinite memory for Claude Code.