Host Integration Upgrades

August 15, 2026 ยท View on GitHub

Updating SuperLocalMemory updates the SLM executable. It does not silently rewrite an AI application's configuration, hooks, plugin state, or workspace instructions. Those files belong to the operator and may contain unrelated tools, policies, or credentials.

Use the consented host-upgrade flow after upgrading through npm, pip, or a repository checkout:

slm upgrade-hosts

This is a read-only preview. It detects existing SLM integrations and reports what is safe to refresh. It does not create an SLM data directory, start a daemon, change a configuration file, or enable the experimental PreToolUse gate.

Apply an upgrade

Review the preview, then target the hosts you approve:

slm upgrade-hosts --host codex --apply
slm upgrade-hosts --host cursor --apply

To refresh every already integrated host in one deliberate action:

slm upgrade-hosts --all-detected --apply

--apply without --host or --all-detected is rejected. This prevents a package update from discovering and changing every supported application.

What is changed

The command uses a merge-not-clobber rule. Existing portable MCP blocks are verified and left intact so host-specific command paths, environment variables, and unrelated MCP servers are not downgraded. For Codex, it may refresh only SLM-owned skills, agents, and lifecycle hooks; the MCP block remains intact.

Claude Code's SLM integration is a marketplace plugin. Refresh it through Claude Code's own plugin manager:

claude plugin update superlocalmemory@qualixar

Then restart the affected host application. Finally verify the local runtime:

slm doctor

First-time setup

For a new machine, use the interactive setup flow instead of the upgrade flow:

slm setup

The wizard detects integrations and asks before connecting them. For a single new host, use slm connect <host>; see IDE Setup for supported hosts and host-specific restart instructions.

Package-manager behavior

The npm package creates a private Python runtime. The pip package installs into the activated Python environment. Neither package manager is used to perform host activation automatically: installation and host consent are separate operations. This separation is intentional and keeps an upgrade reversible, inspectable, and compatible with developer-managed configuration.