Purpose
ML.ai is an AI coding agent for real engineering tasks. Users can hand it bug fixes, test writing, or week-long code migration tasks. It reads the code repository, routes each step of the task to the appropriate model for execution, and shows a diff before any changes are written to disk; it only actually writes after the user clicks "Allow".
Main Capabilities
- Complete tasks end to end: The agent can reproduce bugs, edit code, run test suites, and open a pull request (PR) with a plain-language summary of each change.
- Automatic model routing: Each step is routed to the appropriate model, from fast models that handle boilerplate code to heavyweight models that tackle tricky race conditions. Users set the level of effort and constraints without needing to specify a vendor.
- Parallel agents: Supports running multiple tasks simultaneously, with each agent working in an isolated sandbox and returning diffs for review, compressing weeks of work into days.
- Adapts to how teams build: Teach it team standards once through MLAI.md and reusable Skills, and every subsequent task will automatically apply those standards, requiring less and less supervision as the collaboration grows.
- Always-on background work: Can schedule routine but important work such as issue triage, dependency audits, and CI monitoring. It runs autonomously and notifies only when a user decision is needed.
- Unlimited runs: Uses fixed pricing, with no visible token counter. The agent keeps working until the task is verified, not until the quota runs out.
- Multi-platform access: Can be used in the desktop app, web app, IDE (VS Code, JetBrains) extensions, and the terminal, all connected through one ML.ai account. The terminal version is scriptable, pipeable, and suitable for CI.
How to Use
Users can hand tasks to the agent from the terminal, IDE, GitHub, or the web. For example, by pointing it at an issue, the agent will automatically reproduce the bug, edit code, run the test suite, and open a PR for review. The agent shows the diff, and the user reviews it and decides whether to merge. Nothing is merged automatically.
Applicable Scenarios
- Bug fixes: such as a payment double-charge issue, where the agent reproduces it, edits the code, adds an idempotency key, and covers it with tests.
- Code migration: such as migrating from session cookies to passkeys, where the agent adds WebAuthn registration andAssertions, retain one version of password fallback, and update end-to-end tests.
- Data migration: If using the @migrate-db skill, add fields, backfill data, and generate rollback SQL.
- Routine tasks: If CI monitoring discovers a flaky test, the agent isolates the test, replaces the wait condition, and opens a patch.
Limitations and Considerations
- It is currently launching first in India, with expansion to more regions to follow.
- Fair use and concurrency rules apply.
- The benchmarks measure model capabilities in a fixed test environment and do not guarantee results on users' codebases.
- Benchmark data for other technology stacks comes from third-party leaderboards and vendor reports (as of September 2026), has not been independently verified by ML.ai, and low-confidence cells are estimates.
- On pricing: every plan includes Infinite, and users choose the number of concurrent tasks rather than the number of tokens. The Individual plan supports 2 parallel tasks, the Professional plan supports 6 parallel tasks, and the Team plan supports 20 parallel tasks. The Team plan is billed annually with a minimum of 5 seats. The Enterprise plan offers self-hosted or VPC deployment, bring-your-own model endpoints, data residency and private routing, a dedicated solutions engineer, and more. Specific prices are not yet available.
