Switchy v2: team AI workspaces
By Ash Eryani
Switchy started as a multi-model chat with persistent memory. One person, one inbox, every AI. That shape works, but it has a ceiling: the moment a second person needs the same context, you are back to pasting transcripts into a doc and re-explaining the codebase. Today we are shipping the version without that ceiling.
Switchy v2 is a shared AI workspace for small teams. A Project is a container of scoped memory where your team, the AI models you choose, and the tools you connect (GitHub, Notion, your own MCPs) all share one thread of context. You mention people. You mention models. You mention tools. The AI is silent until you call it. The memory is shared, with three visibilities - private to you, scoped to a Project, or scoped to the whole org - and it shows up everywhere your teammates use AI: Claude Desktop, Cursor, ChatGPT, anything that speaks MCP.
Why this shape and not another? Three reasons. First, AI work is collaborative now in practice but solo in tooling. Two engineers will both ask Claude the same question, separately, and end up with two slightly different answers and no shared record. We wanted the answer once, in front of both of them, with the deciding moment captured. Second, MCP turned the model from a sealed box into something you can plug into. Once that happened, "AI workspace" stopped meaning "chat client" and started meaning "the place your team works with AI." Third, memory only earns its keep when it compounds across people. A team memory is far more valuable than a personal one.
If you work solo, none of this gets in your way. Your account is its own personal organization with one Project called General. Every memory, every chat, every API key lives there. If you only ever use Switchy alone, the team machinery stays out of sight. When you are ready to bring a teammate in, the invite link is one click. Nothing migrates.
For teams, the new plan is $49/month plus $12 per seat ($39 + $10 yearly). It includes 5,000 pooled credits a month plus 1,000 per seat, unlimited Projects, and connections to GitHub, Notion, and any HTTPS MCP server. Solo Pro stays at $19/month. Starter stays free. Plan details at /pricing.
A guided tour of the loop is at /tour. The 5-minute quickstart is at /docs/quickstart/signup. The full migration guide for v0.2 SDK users is at /docs/guides/migrating-from-v1. The v2 API is documented under /docs/api/overview with an OpenAPI 3.1 spec at /api/openapi.json. SDKs are out as @switchy-ai/sdk@0.3.0 (npm) and switchy-sdk==0.3.0 (pypi).
A few things that aren't obvious from the surface but matter to me: AI is silent in multi-participant chats unless you @mention it; memory visibility is enforced at the SQL layer, not in application code; outbound MCP calls go through an SSRF guard that runs on every call (not just at registration) because DNS rebinding is a real attack; org membership doesn't auto-grant Project access (least privilege by default). The threat model is documented per-wave in the rebuild docs.
If Switchy reads to you like one more chat tool, that is the wrong frame - give it ten minutes with one teammate. The shape is different.
Start at switchy.build.