Getting Started

July 23, 2026 · View on GitHub

SuperLocalMemory V3 Documentation https://superlocalmemory.com | Part of Qualixar

Install the CLI, activate the product explicitly, and verify one store/recall round trip.

Product boundary

SLM is useful when you need a user-operated memory service across configured tools:

Local core path by default. Core memory state uses the configured local data root. Optional providers, connectors, backup, model downloads, and skill evolution have separate network behavior and must be enabled or configured.
Named client configurations. MCP and CLI surfaces can point multiple configured tools at one approved data root. Treat a client as verified only when it passes the release integration matrix.
Outcome-aware ranking components. Explicit feedback and qualified outcomes can inform local ranking. Exposure alone is not a positive signal.

SuperLocalMemory is built for one developer, one laptop, many tools. Team / multi-user memory is a different product (SLM-Mesh).

Integration surface: SLM exposes MCP and CLI contracts. Protocol compatibility does not by itself prove install, lifecycle, identity, and cross-client behavior for every product that implements MCP.

Prerequisites

Node.js 18 or later
Python 3.11 or later — macOS ships 3.9; use brew install python@3.11 or a version manager. Ubuntu 22.04 users: sudo add-apt-repository ppa:deadsnakes/ppa && sudo apt install python3.11 python3.11-venv
An AI coding tool (Claude Code, Cursor, VS Code, Windsurf, or any MCP-compatible IDE)

Linux / Ubuntu 22.04: Install in a venv to avoid system-Python conflicts:
python3.11 -m venv ~/.slm-venv && source ~/.slm-venv/bin/activate
python -m pip install superlocalmemory
Then set SLM_PYTHON=~/.slm-venv/bin/python so slm uses that interpreter.

Install

npm install -g superlocalmemory

This installs the slm command globally.

Run the Setup Wizard

slm setup

The wizard walks you through three choices:

Pick your mode
- Mode A (default) — Local core memory path. Optional downloads, connectors, backups, and explicitly enabled integrations can use the network.
- Mode B — Local LLM. Uses Ollama on your machine for smarter recall.
- Mode C — Cloud LLM. Uses OpenAI, Anthropic, or another provider for maximum power.
Connect your IDE — The wizard detects installed IDEs and configures them automatically.
Verify installation — A quick self-test confirms everything works.

Tip: Start with Mode A. You can switch to B or C anytime with slm mode b or slm mode c.

Store Your First Memory

slm remember "The project uses PostgreSQL 16 on port 5433, not the default 5432" --json

You should see:

{"success":true,"command":"remember","data":{"operation_id":"<opaque-operation-id>","materialization_state":"queryable","fact_ids":["<queryable-fact-id>"],"note":"queryable now; canonical enrichment pending"}}

The exact identifiers differ on every installation. Use --sync if your next step requires complete rather than the default queryable-first receipt.

Recall a Memory

slm recall "what database port do we use"

Output:

[1] The project uses PostgreSQL 16 on port 5433, not the default 5432
    Relevance: 0.94 | Stored: 2 minutes ago | Profile: default

The value is query-relative relevance, not answer confidence. V3.7 declares calibration_status: "uncalibrated" and answer_confidence: null; see the retrieval score contract.

Check System Status

slm status

This shows:

Current mode (A, B, or C)
Active profile
Total memories stored
Database location
Health of math layers (Fisher, Sheaf, Langevin)

How It Works With Your IDE

Automation depends on the client plus the hooks/instructions you explicitly enable:

Auto-recall — Supported session hooks can request bounded, untrusted evidence context.
Auto-capture — Supported observe hooks can submit content to configured admission rules.

You can still use slm remember and slm recall from the terminal whenever you want explicit control.

Try Bounded Loops (v3.8.0)

A bounded loop runs laps until an independent gate passes — not until the agent claims it is done. Every lap is persisted to your SLM data root, queryable via slm recall, and visible on the dashboard.

Verify the engine end to end with the built-in demo:

slm loop demo

Expected output:

✓ [DONE] gate passed on lap 3 (laps: 3)
   lap 1: changed  gate-fail  demo gate: lap 1
   lap 2: changed  gate-fail  demo gate: lap 2
   lap 3: changed  gate-pass  demo gate: lap 3
run_id: <id>  (recall with tag loop:convergence-demo)

Then recall the stored lap history:

slm recall "convergence-demo loop"

For the full parameter set, see CLI Reference → Bounded Loops and MCP Tools Reference → Bounded-Loop Tools.

Next Steps

What you want to do	Guide
Set up a specific IDE	IDE Setup
Switch modes or providers	Configuration
Learn all CLI commands	CLI Reference
Migrate from V2	Migration from V2
Understand how it works	Architecture
Use SLM from a Python framework	Framework Adapters