V3.6 — Configuration

Three modes. One for every setup.

SuperLocalMemory adapts to your infrastructure — from zero-cloud local installs to cloud-LLM power mode. Pick once, change any time.

DEFAULT · RECOMMENDED
A

Fully local.
No LLM required.

Mode A uses TF-IDF statistical embeddings — no neural network, no GPU, no external API. All storage lives in SQLite on your machine. Retrieval uses the Fisher-Rao information metric for mathematically-grounded recall, not cosine heuristics.

  • EmbeddingTF-IDF (statistical, local)
  • StorageSQLite + in-memory index
  • GPUNot required
  • InternetNever
  • Setup time< 1 minute
  • Accuracy (LoCoMo)74.8%
Use when
  • Privacy-first orgs
  • CI/CD pipelines
  • Low-resource machines
  • Air-gapped environments
setup — Mode A
# Install SuperLocalMemory $ npm install -g superlocalmemory
# Setup with Mode A (default) $ slm setup --mode A
✓ Mode A active · TF-IDF engine ready ✓ SQLite initialised · 0 facts indexed ✓ Fisher-Rao retrieval · cache: on
# Verify status $ slm status engine: tfidf · mode: A · internet: never ✓ Ready — no cloud, no GPU, no API key
LOCAL LLM
B

Semantic accuracy.
Zero cloud.

Mode B plugs into your local Ollama instance to generate high-quality neural embeddings. Same Fisher-Rao retrieval, same local storage — but with the semantic depth of a full language model running on your hardware.

  • EmbeddingOllama (local neural)
  • StorageSQLite + LanceDB vector
  • GPURecommended (4GB VRAM+)
  • InternetNever
  • Setup time~5 minutes (Ollama install)
  • Accuracy (LoCoMo)82.3%
Use when
  • Max privacy + max accuracy
  • Developers with local GPU
  • Teams already running Ollama
setup — Mode B
# Pull an embedding model via Ollama first $ ollama pull nomic-embed-text pulling nomic-embed-text:latest … done
# Setup SLM with Mode B $ slm setup --mode B
✓ Ollama detected · nomic-embed-text ready ✓ LanceDB initialised · vector index: on ✓ Fisher-Rao retrieval · cache: on
# Test semantic recall $ slm recall "token expiry" ↳ JWT, refreshed every 15m [conf 0.99 · neural]
CLOUD
C

Maximum accuracy.
API-powered.

Mode C uses OpenAI or Anthropic embedding APIs for the highest-quality semantic representations. Your memories stay local — only embedding generation hits the cloud. Ideal when accuracy matters more than absolute privacy.

  • EmbeddingOpenAI text-embedding-3 / Anthropic
  • StorageSQLite + LanceDB vector
  • GPUNot required
  • InternetEmbedding calls only
  • Setup time~2 minutes (API key)
  • Accuracy (LoCoMo)87.7%
Use when
  • Maximum recall accuracy
  • Teams with existing OpenAI / Anthropic contracts
setup — Mode C
# Export your API key (never stored by SLM) $ export OPENAI_API_KEY=sk-...
# Setup with Mode C — OpenAI provider $ slm setup --mode C --provider openai
✓ Provider: openai · text-embedding-3-small ✓ LanceDB initialised · vector index: on ✓ Memories stay local · only embeds hit API
# Or switch to Anthropic $ slm setup --mode C --provider anthropic ✓ Mode C active · 87.7% LoCoMo accuracy
Side by Side

Compare the three modes.

Every tradeoff, visible at a glance. No asterisks.

Feature Mode A Mode B Mode C
No internet required ✓ ✓ –
No GPU needed ✓ – ✓
No API key ✓ ✓ –
Setup time < 1 min ~5 min ~2 min
LoCoMo accuracy 74.8% 82.3% 87.7%
Embedding type TF-IDF (statistical) Ollama (neural, local) OpenAI / Anthropic
Storage backend SQLite + in-memory SQLite + LanceDB SQLite + LanceDB
Recommended for Privacy · CI/CD · Default Local GPU teams Max accuracy
Zero lock-in

Change mode any time — your memories persist.

SLM re-indexes your existing memories using the new embedding method in the background. No data loss. No interruption. The command takes under a second — re-indexing happens asynchronously.

mode switch
# Switch from Mode A to Mode B $ slm mode --set B
✓ Mode changed: A → B · 528 facts preserved ✓ Re-indexing in background (Ollama neural)
# Check progress any time $ slm status mode: B · indexed: 528/528 · engine: ollama ✓ Re-index complete · all memories available
Common Questions

Questions about modes.