Cobalt Code is available today at cobaltcode.ai. It is a software factory for product builders: connect a repository, describe the outcome you need, and Cobalt gives an AI developer an isolated cloud computer where it can build, run, test, and preview the real application.
That distinction matters. Cobalt is not just a coding agent or a chat interface that returns suggestions. It manages the environment around the agent: the repository connection, computer lifecycle, long-running services, live previews, task history, reusable starting points, and delivery through Git.
From a request to running software
Start with the kind of task you would hand to another developer: add a feature, reproduce a bug, test a refactor, or turn an idea into a working prototype. Cobalt provisions a dedicated computer for the task and prepares the selected repository.
Inside that environment, the AI developer can:
- install the project's actual dependencies;
- edit code across the repository;
- run tests and diagnose failures;
- start databases, background services, and Docker workloads;
- launch the application and expose authenticated live previews; and
- prepare commits and pull requests for review.
The result is not limited to a patch. You can inspect the running application beside the conversation, steer the work with follow-up messages, review the test results, and verify the Git changes before anything is merged.
A computer that belongs to the task
Each task gets an isolated, persistent computer. The computer keeps working if the browser disconnects, and the task can be resumed later with its files, services, agent session, and history intact.
This makes Cobalt useful for work that does not fit into a short request-response loop. A full-stack application can keep its database and supporting services running. A difficult bug can retain the exact state in which it was reproduced. Several features or experiments can progress in parallel without competing for one local checkout.
When a computer becomes a useful development environment, it can be saved as a reusable starting point. Future tasks can begin with dependencies installed, services configured, and project state ready to use, while still receiving their own independent computer.
Built for the product development loop
Cobalt brings implementation and review into one workspace. Product builders can describe an outcome in plain language, watch the application change through a live preview, and respond to what they can actually use rather than debating a static mockup or an unverified diff.
Developers retain the controls that matter. Work stays connected to the repository, test and command activity remains visible, and completed changes return through the Git workflow the team already uses.
Repository and agent credentials are provided only when work needs them. They are not stored in task snapshots, and each task remains bounded to its own computer.
Watch the demo
See a repository set up on a persistent computer, tests and PostgreSQL running, the live application preview, and the finished work prepared for delivery through Git.
Watch the Cobalt Code demo on YouTube.
Start a task
Visit cobaltcode.ai to learn more and start your first task.
