ML.ai
ML.ai helps developers assign repository work to a coding agent. Its documented editor module reads project files, runs commands and proposes changes for approval. The product belongs to Pixis, and its inference system chooses the agent and routing tier for a task.
Repository work with a review step
The feature documentation describes several agent roles, including General, Explore, Architect and Plan. General carries out changes end to end; the other roles give users ways to scope or investigate work before authorizing edits. Plan mode denies edit and write tools and applies a stricter command budget. That separation is useful when you want an implementation plan without changing the repository during the discussion.
ML.ai says it classifies shell commands by their effects before execution. Its permission interface offers Allow, Always or Deny, with a diff before the user decides. A denial goes back to the model with a reason. Those are product controls, not a substitute for reviewing a proposed change or maintaining a test suite.
Background jobs and reusable instructions
A delegated job runs in a separate session and can continue when its panel closes or reloads. The main conversation remains available while the job runs. For repeatable work, users can attach packaged, project or global skills through the slash menu.
The product also supports MCP servers scoped to a project or configured globally. The documented connection options include environment variables, headers, timeouts and OAuth sign-in. This gives a team a way to expose its own tools to the agent, with the same need to check permissions and access that applies to any connected service.
Model routing and product availability
The introduction describes two routing tiers, ML.ai Standard and ML.ai High, and says users do not choose the underlying model. It identifies the VS Code and Cursor coding extension as shipped, while marking CLI, Market and Voice modules as parked.
Availability needs care here: the homepage promotes desktop, web, IDE and terminal access and describes a private beta launching in India first. That is broader than the documentation's shipped-module list. Confirm the interface you need before planning a CLI or desktop deployment.
Pricing needs confirmation
The homepage advertises flat subscriptions with concurrency and fair-use limits. The separate pricing page describes monthly plans with included execution usage and fees beyond that allowance. Those terms differ. Treat the current checkout or a written vendor quote as the buying decision, rather than assuming the homepage's unlimited-work language applies to every product or account. Evaluate proposed changes against your own test suite before relying on advertised benchmarks.