Capy launches a multiplayer cloud coding platform with 10-agent orchestration

• Capy launched what it calls a "multiplayer cloud coding platform," packaging planning, building, testing, review, PR handling, Slack, Linear, and "computer use" into one hosted workspace launch post.
• The core execution model is parallel orchestration: Capy's Captain demo says its Captain agent can run "10+ build agents at once" in "self-healing loops" rather than a single local-agent session.
• Capy is pitching a full async dev loop, with a review agent demo for bug-catching, a Linux VM demo for autonomous testing on a Linux VM, and an PR view demo "AI-native PR view" for code review.
• The workspace also appears designed as a model-agnostic front end: Capy's model support post says it supports GPT, Claude, Kimi, and Grok, while subscription post says users can "bring your Codex or Copilot subs."

Capy's launch pitch is a hosted coding environment, not another editor plugin. In the company's launch post, it describes a system that "plans, builds, tests, and reviews your code" and produces "self-healing PRs fully async," with PR view, review agent, computer use, Slack, Linear, and ChatGPT subscription support folded into one tool.

The product page linked from the site post frames that as "the IDE for the parallel age" and a dashboard for orchestrating multiple coding agents, including task management, code review, and deployment from a single cloud workspace. That matters because the implementation center of gravity appears to be the hosted environment itself, not a local CLI loop with occasional remote runs.

Capy's clearest technical claim is orchestration. In the Captain demo, the company says Captain can coordinate "10+ build agents at once in self-healing loops," suggesting concurrent attempts, retries, or subtask decomposition inside one sprint-sized workflow rather than one model call at a time.

The rest of the thread fills in that loop. Capy's Linux VM demo says the platform "autonomously tests code on its Linux VM," while the review agent demo says a separate review agent works with Captain "to catch bugs." Combined with the launch claim of "self-healing PRs" in the main announcement, the workflow looks like build agents generating changes, a hosted VM running tests, and a review layer feeding failures back into another repair cycle before a PR is surfaced.

Capy is also positioning the product as a control plane across models and existing subscriptions. The model support post says it works with "leading models like GPT, Claude, Kimi, Grok," and the subscription post adds that teams can "bring your Codex or Copilot subs" instead of using only one bundled model path.

On the developer-facing side, Capy pairs that model layer with an "AI-native PR view," as shown in the PR demo, which implies the handoff artifact is still a pull request even if most execution happens in cloud agents. The result is a hosted workflow that bundles agent orchestration, test execution, review, and PR inspection into one interface, with model choice treated as a pluggable backend rather than the product itself.