One memory. Every model.
Decisions, facts, and context — surfaced the moment anyone needs them. No copy-paste. No re-prompting.
The thread doesn’t restart on Monday.
Every conversation leaves something behind: a decision, a number, a constraint, a link. Switchy keeps that thread alive — and surfaces the right piece the next time anyone needs it.
Three scopes. No mystery about who sees what.
Every memory is tagged with one of three scopes when it’s saved. You can change it later. You can always see the tag.
Just you.
A private memory is invisible to your teammates and your AI models — until you change the scope. Use it for half-formed ideas, drafts, or anything you’d write in a personal notes app.
Everyone in this Space.
The default for team work. Decisions, facts, and sources visible to every member of this Space — and to every AI mentioned in it. The shared ground state of one project.
Everyone in your org.
For things that should follow everyone everywhere — brand, security, billing constraints, holiday schedule. Visible across every Space, every teammate, every model your org uses.
Your Space memory is an MCP server.
Paste one URL into your editor’s MCP config. The same context the team built up this morning is in your prompt this afternoon — inside Claude Code, Cursor, Claude Desktop, ChatGPT, anywhere MCP is supported.
Switching models doesn’t reset memory.
@claude one minute, @gpt-5 the next, @gemini after that — every model in the Space sees the same context. The thread is the source of truth, not the provider.
Claude
Gemini
Mistral
GrokYou see everything that’s stored. You can delete anything.
Private stays private.
Memories saved as PRIVATE are visible only to you — not your teammates, not the AI models in your Space. The lock label is on every memory in the UI.
Browse what’s stored.
The Memory view lists every saved item, when it was saved, by whom, and what scope it has. Filter by Space, by author, by date. No black box.
Delete with one click.
Remove a single memory, a whole conversation, or an entire Space’s history. Deletes propagate to every surface — the app, the API, and the MCP server.
Build your own retrieval. Build your own UI. Keep memory in your stack.
Read and write a Space’s memory directly from your code. The same API powers the app, the MCP server, and your custom integrations — one source of truth, three surfaces.
Query memory with a question; write memory with a fact, decision, or constraint. Scope tags travel through the API the same way they do in the app — what’s PRIVATE stays PRIVATE.
Bearer auth, JSON in, JSON out. Stream replies if you want; batch them if you don’t.
Read the docs// npm i @switchy/sdk
import { Switchy } from "@switchy/sdk";
const sx = new Switchy({ space: "team-product" });
// Recall
const hits = await sx.memory.recall({
query: "how did we decide onboarding step 3?",
scope: ["project", "org"],
});
// Save
await sx.memory.save({
text: "Default model is Claude Sonnet 4.5",
scope: "org",
kind: "decision",
});