Run coding, review and testing agents in isolated cloud environments from Capy, Slack, Linear or developer integrations.
Capy is a cloud software-engineering platform for running AI coding agents across repositories and team workflows. It includes Captain, Build and Review agents, with a web interface for organising parallel tasks. Each task runs in a development environment where the agent can inspect files, edit code and execute checks rather than only suggest a patch in conversation.
Teams can use Capy for bug triage, migrations, backlog work and pull-request review. Its native review interface brings proposed changes and review feedback into the platform. Browser and computer-use capabilities support end-to-end testing, including visual evidence from an agent run. Generated changes still need the same functional and security review expected of other contributions to a repository.
The service supports several model providers and lets users choose a model combination for the task. Compatible existing subscriptions, including ChatGPT Codex, GitHub Copilot and SuperGrok, can be connected for supported models. Enterprise arrangements add bring-your-own-key options. Provider availability and the subscription’s own limits continue to matter.
Automations can start from schedules, Slack messages, Linear updates, GitHub events or API and webhook calls. Every run has its own task history and configured environment. Limits can bound how often an automation runs, while durable files retain relevant working material across runs. Failed tasks remain inspectable so a team can correct the instructions or environment before trying again.
Pro starts with a monthly usage budget, and larger tiers add more included usage. Credits cover model costs and virtual-machine compute; unused allowance does not roll over. Additional credits and optional automatic top-ups are available. Enterprise controls include identity and audit options. A safe rollout begins with a bounded repository task, checks the resulting diff and test evidence, and only then enables recurring triggers. Parallel execution expands capacity, but it should not broaden repository permissions or bypass the team’s normal approval process.
