Two engineers, one live session, one agent on one running drive. Stop trading a screen-share where one person drives and the other watches. *Both of you* prompt the same agent, on the same files and terminal, at the same time.
One of you drives, the other watches a video of your screen.
Both of you prompt one agent in one live session, together.
Validation runs on blur for each field. Errors render in a toast for now; tell me if you want them inline under the field.
Matched the login form’s FieldError component. Errors sit under each field now.
Pay is disabled until the form validates. 9 tests pass, including the three new ones for the error states.
bun test after every step, load test before the flag movesProtect /v2 from bursty API keys without slowing normal traffic. One key should never be able to starve the others.
A token bucket per API key, 100 requests per minute, stored in Redis so all three replicas share state. Over-limit requests get a 429 with a Retry-After header. Ship behind the rate_limit_v2 flag and roll out gradually.
bun test after every step. Load test with 1k keys before the flag moves to 50%. Watch 429 rate and p99 in the dashboard for a day at 10%.
Not a screen-share and not two checkouts that merge later. One shared session on one running drive, with everything you need to work side by side. Every line below is running in production today.
No driver and no watcher. Two engineers prompt one agent in one live session, on one running filesystem and terminal, and watch the same work land. Not two copies of the codebase trying to merge later.
See who is in the session and what they are looking at. Follow a teammate's view to ride along with their train of thought, and use the in-session chat overlay to talk without leaving the work.
The terminal and the integrated localhost browser are shared and auto-reloading, so you both see the same output and the same preview. Inspect an element, leave a comment, and send it straight to the agent.
Keep a session private while you sketch, share it with your pairing partner by name, or open it to the whole drive. Per-session access control means you decide the audience, session by session.
A drive stays warm in the cloud with your files, terminal, and dependencies ready. The agent reads the repo once for the session and holds it warm, so the two of you work against the same live state instead of each paying to load context on your own. Use 70+ models on your own keys, scoped per drive, and push to your real git when you are done. It sleeps when idle, and idle time is never billed.
On a screen-share, one person types and everyone else watches a video of it. In Coshell you are both in the same live session, prompting the same agent on the same running drive. Your partner is not watching a stream of your machine, they are working in the environment with you, with their own prompts, presence, and follow.
You are prompting one shared agent on one running environment, not two copies that merge later, so there are no diverging changes to reconcile afterward. If work runs concurrently and two edits reach for the same lines, an overlap layer catches it before they clobber. The repo is read in once and stays warm for the session, so you are both working against the same live state.
Yes. Coshell works from any browser and from the CLI. Your partner can device-login and attach to the drive by name, and the stock OpenCode CLI attaches unchanged. You can be in the browser and they can be in the terminal, both in the same live session on the same drive.
Open a drive in the browser, or attach from the CLI. Bring your pairing partner into the same live session and prompt one agent together, on the same running drive.