Configuration

July 16, 2026 · View on GitHub

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

Control how SuperLocalMemory stores, retrieves, and processes your memories.

Three Operating Modes

For the published LoCoMo result scopes behind Modes A and C, see Benchmark Evidence. Mode B has no separately published LoCoMo run.

SuperLocalMemory runs in one of three modes. You pick the trade-off between privacy and power.

Mode	What it does	Needs API key?	Data leaves your machine?
A: Local	Retrieval without a model-provider call in the core path.	No	Optional integrations may transmit data
B: Local LLM	Mode A + a local LLM via Ollama.	No	Depends on the Ollama endpoint and optional integrations
C: Cloud LLM	Mode B + configured cloud-provider enrichment and/or answer construction.	Yes	Configured query, ingestion, or enrichment content may be sent

Check your current mode

slm mode

Switch modes

slm mode a    # Zero-cloud (default)
slm mode b    # Local LLM
slm mode c    # Cloud LLM

Switching modes takes effect immediately. No data is lost.

Mode A: Zero-Cloud (Default)

Core memory operations run against the local data root. Optional model and dependency downloads, connectors, backup, and other enabled integrations can use the network.

Best for: deployments that want a local core path and can govern optional integrations explicitly. Regulatory compliance still requires deployment-specific assessment.

Mode B: Local LLM

Everything from Mode A, plus a local LLM (via Ollama) that improves recall by understanding query intent and reranking results.

Setup:

# Install Ollama using its reviewed package/instructions for your platform.
# macOS example:
brew install ollama

# Pull a model
ollama pull llama3.2

# Switch to Mode B
slm mode b

Best for: developers who can operate the selected local model and separately govern optional networked integrations.

Mode C: Cloud LLM

Everything from Mode B, plus a cloud LLM for cross-encoder reranking and agentic multi-round retrieval. Highest recall quality.

Setup:

slm mode c
slm provider set openai

You will be prompted for your API key (stored locally in your config file, never transmitted except to the provider you choose).

Best for: deployments that have approved the configured provider data path.

Provider Configuration

Mode C supports multiple LLM providers.

Set your provider

slm provider           # Show current provider
slm provider set       # Interactive provider selector

Supported providers

Provider	Command	Env variable
OpenAI	`slm provider set openai`	`OPENAI_API_KEY`
Anthropic	`slm provider set anthropic`	`ANTHROPIC_API_KEY`
Azure OpenAI	`slm provider set azure`	`AZURE_OPENAI_API_KEY`
Ollama (local)	`slm provider set ollama`	None needed
OpenRouter	`slm provider set openrouter`	`OPENROUTER_API_KEY`

Set API keys

You can set keys interactively or via environment variables:

# Interactive (stored in config file)
slm provider set openai
# Prompts: Enter your OpenAI API key: sk-...

# Via environment variable (takes precedence)
export OPENAI_API_KEY="sk-..."

Config File

All settings live in:

~/.superlocalmemory/config.json

Example config

{
  "mode": "a",
  "profile": "default",
  "provider": {
    "name": "openai",
    "model": "gpt-4o-mini",
    "api_key_env": "OPENAI_API_KEY"
  },
  "auto_capture": true,
  "auto_recall": true,
  "embedding_model": "all-MiniLM-L6-v2",
  "max_recall_results": 10,
  "retention": {
    "default_policy": "indefinite"
  }
}

Key settings

Setting	Default	Description
`mode`	`"a"`	Operating mode: `a`, `b`, or `c`
`profile`	`"default"`	Active memory profile
`auto_capture`	`true`	Automatically store decisions and context
`auto_recall`	`true`	Automatically inject relevant memories
`embedding_model`	`"all-MiniLM-L6-v2"`	Sentence transformer for semantic search
`max_recall_results`	`10`	Maximum memories returned per query

Environment Variables

These override config file settings when set:

Variable	Purpose
`SLM_MODE`	Override operating mode
`SLM_PROFILE`	Override active profile
`SLM_DATA_DIR`	Override data directory (default: `~/.superlocalmemory/`)
`OPENAI_API_KEY`	OpenAI API key for Mode C
`ANTHROPIC_API_KEY`	Anthropic API key for Mode C
`AZURE_OPENAI_API_KEY`	Azure OpenAI API key for Mode C
`OPENROUTER_API_KEY`	OpenRouter API key for Mode C

Database Location

All data is stored locally in:

~/.superlocalmemory/memory.db    # SQLite database
~/.superlocalmemory/config.json  # Configuration
~/.superlocalmemory/backups/     # Automatic backups

To use a custom location:

export SLM_DATA_DIR="/path/to/your/data"

Multi-Machine Mesh (v3.4.48+)

Variable	Default	Description
`SLM_MESH_PEER_URL`	unset	Full URL of remote SLM instance (e.g., `http://192.168.1.100:8765`)
`SLM_MESH_SHARED_SECRET`	unset	Shared bearer token — same on both machines. Required when `SLM_MESH_HOST` is not localhost.
`SLM_MESH_HOST`	`127.0.0.1`	IP to bind this machine's mesh listener
`SLM_MESH_WS_PORT`	`7900`	Port used for mDNS service announcement
`SLM_MESH_DISCOVERY`	`on`	Set to `off` to disable mDNS auto-discovery

See Multi-Machine Setup for full setup guide.

SLM v3.6 adds the Optimize module — Cache + Compress + Align for LLM cost reduction. Configuration lives in a separate file at ~/.superlocalmemory/optimize.json and hot-reloads within 2 seconds — no daemon restart required.

Master Switches

Setting	Default	Description
`optimize enabled`	`true`	Master ON/OFF
`cache enabled`	`true`	Cache lookups (exact match)
`semantic cache`	`false`	vCache semantic (opt-in)
`compression`	`safe`	`safe` (lossless) or `aggressive` (lossy prose allowed)

Quick Toggle via CLI

slm optimize on                     # Enable all
slm optimize off                    # Disable all
slm cache semantic on               # Enable semantic cache
slm compress mode aggressive        # Enable aggressive compression

Config File Location

~/.superlocalmemory/optimize.json   # Written by UI, CLI, and API

Hot-Reload

The daemon polls this file every 2 seconds. On change, all settings take effect without restart. Config version is auto-incremented for change tracking.

Full Reference

See docs/optimize-config.md for all 45+ config fields with defaults and descriptions.