-
-
Notifications
You must be signed in to change notification settings - Fork 34
Host Upgrades
An SLM package update upgrades SuperLocalMemory itself. It never silently rewrites Codex, Claude Code, Cursor, VS Code, Antigravity, Hermes, or another host's configuration.
After upgrading through npm, pip, or a repository checkout, inspect your local integrations:
slm upgrade-hostsThe default command is a read-only preview. It does not create memory data, start the daemon, edit host configuration, or turn on the experimental tool gate.
To apply a reviewed change, name the host explicitly:
slm upgrade-hosts --host codex --applyOr deliberately refresh all hosts that already contain an SLM integration:
slm upgrade-hosts --all-detected --applyThe command rejects an unbounded --apply. Existing portable MCP entries are
verified and preserved rather than replaced with a generic configuration. This
protects custom command paths, agent identities, and unrelated MCP servers.
Codex may refresh its SLM-owned skills, agents, and lifecycle hooks. Claude Code uses its own marketplace plugin lifecycle:
claude plugin update superlocalmemory@qualixarRestart the relevant application, then run:
slm doctorFor first-time setup, use slm setup; it detects integrations and asks before
connecting them. Read the full source documentation in
Host Integration Upgrades.
SuperLocalMemory V4.0.3 — Local-first memory with explicit data-path controls. Current docs: Home · Installation · CLI · MCP whole 91 (full 46) · FAQ
Part of Qualixar | Created by Varun Pratap Bhardwaj | GitHub · CHANGELOG
Platform boundary: Apple Silicon macOS · 64-bit Windows · 64-bit Linux — Intel Mac and 32-bit Windows not supported (
cryptography==50.0.0).
SuperLocalMemory V4.0.3
Getting Started
- Installation
- Quick Start Tutorial
- Getting Started
- Modes Explained
- IDE Setup
- Host Integration Upgrades
- Migration from V2
Reference
- CLI Commands
- MCP Tools — 91 whole / 58 power / 46 full / 28 code / 14 core
- Retrieval Score Contract
- Auto-Memory
- Active Memory (V3.1)
Integrations
- Framework Adapters — 9 adapters
- Bounded Loops
- Multi-Agent Memory
Architecture
- Architecture Overview — historical V3, carried into V4
- Capabilities and Operations
- Published Benchmarks — V3 LoCoMo, not a V4 rerun
- Mathematical Foundations — historical V3
-
V4 Reliability Contract — 2,200/2,200, protocol
benchmark/run_all.py --trials 200
Enterprise and Teams
V2 Documentation