Skip to main content
Getting your agent into a Google Meet is the start. What makes it last is the layer underneath — and it’s being built in the open.

​
One identity, everywhere

The hard part isn’t a bot in a Meet — it’s the control plane: one agent identity, scoped admission, and a signed receipt for every call, the same principal across Google Meet, Slack, and open, portable rails. That layer converges on one MCP. The slashvibe-mcp server — presence, DMs, identity — is the network these calls run on; the conferencing tools are folding into it, so eventually you connect /vibe once and just say “invite my agent to this Meet.”
Honest status: today the calls run through a companion conferencing MCP, and the unified install is in progress. We converge, we don’t fork.

​
On the roadmap

The doorbell

Summon a hosted agent into a Meet when someone you’ve authorized sends the code — without your laptop being awake. The summon path is built but dormant behind a feature flag while the desktop doorway remains the launch focus.

Cross-surface

The same agent, the same identity, in Slack huddles and open rooms — not just Google Meet.

The buddy list

A presence + registry layer so you can pull any agent you’re connected to into any room, with consent.

A phone client

A call is the same room whether you’re at a desk or walking around. That’s a client problem, not an architecture one.

​
What holds it together

Not the demo — a weekend clone can copy the emoji and the voice. What compounds is the hard last mile: social turn-taking that reads as a colleague, cross-surface authority and receipts, the operational scars of real rooms, and the habit of teams who already summon their agent. That’s the moat, and it’s earned by running.