Early access · Fall 2026
Wheelhouse puts your team's coding agents — Claude Code, Codex, Gemini, whatever you run — in one shared room. Watch a run live, take the wheel, pause it before it eats the auth layer, or fork it and prove your point. One driver at a time. A log of who did what.
The state of the art, 2026
Here's how teams share agent sessions today: someone runs Claude Code, something interesting happens, they screenshot it into Slack, and four people reply "wait — tell it to…" at a terminal only one of them can touch.
Google Docs fixed this for writing. Figma fixed it for design. Agent sessions are still single-player, which is a little embarrassing for all of us.
What you actually get
One driver at a time — an actual lock, not a vibe. Everyone else watches the run live. Want control? Request the wheel and the driver hands it over in one click. "Assign this run to Priya" is a button, not a Slack thread, and the session log remembers every handoff.
Anyone in the room can pause a run. Not kill — pause. The agent stops, the room decides, the driver resumes or redirects. Because the second-worst thing an agent can do is go wrong, and the worst is go wrong for forty more minutes while you draft a polite DM.
Disagree with the direction? Fork the session mid-run, take your branch, and merge back the parts that worked. Arguments become diffs.
No favorites
Every agent vendor will happily sell you a room — with their agent bolted to the floor. We don't have a model to push.
Bring your own subscriptions, keep your own keys, run them side by side in the same session — in the browser, on any OS. We sell the room, not the seat.
Why we're building this
I run agents all day. The best moments — an agent mid-refactor doing something clever, or something catastrophic — are exactly the moments I want a teammate looking over my shoulder. And the tooling for that, today, is "come look at my screen."
Presence is the easy half, and plenty of tools have it. The hard half is control: who's driving, who can stop it, what happened and when. That's the half we're building first.
It's early. We're looking for a handful of teams to build this with — teams that run agents daily and argue about them weekly. If that's you, get on the list and be loud with us.
— Ben
Questions we keep getting
This fall, in small batches, to teams who'll actually use it and yell at us about it. The optional "what do you run" field genuinely affects ordering.
Claude Code and Codex first, Gemini right behind. If your team runs something else, say so when you sign up — supporting whatever you already use is the whole point.
Free during early access. After that: per-seat, and boring. Watching will always be free — you only ever pay for driving.
There are tools with presence — see who's online, prompt together. Nobody has shipped real control semantics: a driver lock, a governed interrupt, forking with merge. That's the hard part, and it's the part we're building first.
Runs execute in isolated sandboxes; your repo and API keys stay yours. Every prompt, handoff, and interrupt lands in an audit log that belongs to you and exports cleanly.
Early access opens this fall. Small batches, real teams first.