ChatGPT on the web reaches the memory on your computer through optional Web access: sign in with GitHub, choose read-only or read-and-save, and ChatGPT recalls from the same governed store your local agents use. The database never leaves your computer. Needs a paid ChatGPT plan.
Web access is optional and managed from the dashboard’s Connected apps page. Your computer keeps one outbound link, with no open port; a gateway problem never stops local use.
Web results do pass through the gateway and the AI service you connected, so keeping the database local does not mean nothing leaves the computer.
Install SLM first, then follow the ChatGPT steps in the host guide. The menu names are ChatGPT’s and can change.
Use pipx or uv with Python 3.12–3.14. Mode A does not require a model-provider key for its core path. Run setup to initialise the local database and configuration.
pipx install superlocalmemory slm setup
In the dashboard open Connected apps, turn on Web access and add an app. Then in ChatGPT on the web choose Plugins, Add, Add custom MCP server; enter a name and the server URL, set Authentication to OAuth, and put the scopes slm:read, slm:write and slm:session in Base scopes.
Sign in with GitHub on the approval page and tick Allow saving memories if you want ChatGPT to save as well as read. ChatGPT keeps the scopes it saw when the plugin was created, so a plugin created read-only stays read-only: remove it and create it again.
ChatGPT host guide → · Web access guide → · ChatGPT dots use the plugins you connected in ChatGPT.
In a chat, open the tools picker and enable the plugin, then ask ChatGPT to call get_status with the SuperLocalMemory plugin. Your computer must be online; Web access has a free daily allowance per connection, and SLM works fully without it. ChatGPT can also join SLM-Mesh and message the agents on your laptop by name, with your permission.
slm status slm doctor
Once connected, ChatGPT can recall and, if allowed, save through the same governed paths. Available tools depend on the selected MCP profile and the installed SLM release.
Store facts, decisions, and context during a conversation. Recalled context in later sessions is bounded — inspect scope and provenance rather than treating it as a complete restore.
The same configured SLM profile can be reached from ChatGPT, Claude on the web, Muse, Claude Code, Cursor and other MCP clients. Each client still needs its own integration verification.
SLM stores core memory in a local SQLite database in Mode A. ChatGPT, optional providers, and any optional SLM integrations retain their own network behavior.
Hybrid search, fusion and the answer check: ChatGPT hears a supported answer, the closest memories, or “That isn’t in SuperLocalMemory.” Retrieved context is evidence, not verified fact.
Mode A keeps the SLM core memory-content path local. Review the network behavior of the full deployment — ChatGPT, optional providers, connectors, backups, and downloads require separate assessment.
SuperLocalMemory is open source under AGPL v3. Full source code is publicly available. Verify the installed release, optional dependencies, and configured providers.
Install SLM, turn on Web access, connect ChatGPT, then verify with get_status.
Read the guidesHost guides · Web access · Web agent instructions