Magnet

Magnet is a workspace for planning and building software with coding agents. Its current website emphasizes issue definition, reusable project context and reviewing generated changes. The linked documentation describes a shared environment containing issues, specifications, pages, agent sessions and pull requests. It is aimed at teams that want to coordinate coding tasks and inspect the resulting changes in one place.

The documented workflow starts with a task and, optionally, a connected repository. An agent runs in a cloud sandbox, works on the code and streams progress back to the workspace. The documentation describes repository cloning, setup scripts, development servers and pull-request creation. Teams can then review a diff before deciding what to merge. These are vendor-described capabilities; no repository was connected and no agent session was run during this review.

The homepage names tools including Claude Code and Codex, while the cloud page presents APIs and SDKs for hosted agent environments, previews and session management. Compatibility and operational limits were not independently tested. The cloud offering is explicitly marked Early Access and invites waitlist requests. Public login and signup forms are accessible, but their presence does not establish that every user can immediately obtain cloud access.

The reviewed public pages did not establish a price, currency, compute allowance or model-usage entitlement. Repository access and agent execution can involve sensitive source code and consequential write permissions. Before connecting a project, establish the applicable access scope, data retention, model-processing terms and billing conditions. This listing does not verify those protections or promise that autonomous changes will be correct. Generated code still requires appropriate review and testing before deployment.