Skip to content

Security

Control before autonomy.

To200 earns autonomy one gate at a time: scoped access, branch-only fixes, explicit approvals, and evidence attached to every change.

manualassistedauto-PRautopilot

Branch only: every change To200 makes reaches your codebase as a pull request.

To200 commits to its own branch and opens a pull request from it. There is no path that skips the pull request, so branch protection, required reviews, and CI apply to its changes exactly as they apply to your engineers’.

Human review stays between the PR and main by default; Scale and High volume owners can deliberately enable merge-to-main, while branch protection, CI rules, and required reviews still apply.

Four autonomy levels. One set of guardrails.

  1. 01

    Manual

    To200 analyzes and waits. Every stage advances only when a human moves it.

  2. 02

    Assisted

    Analysis and planning run on their own. Fixes start after a human sign-off.

  3. 03

    Auto-PR

    Available on Pro. Checked branch-only fixes can become pull requests automatically while merging stays human by default.

  4. 04

    Autopilot

    Available on Scale and High volume. More stages can advance inside the same branch, evidence, and review guardrails.

Free includes Manual and Assisted control. Pro unlocks Auto-PR; Scale and High volume unlock Autopilot. Raising the level never removes the evidence trail or your repository rules.

Six controls behind every repair.

Branch-based repairs

Fixes are written on branches and move through pull requests. No silent writes to your default branch.

Scoped access

OAuth connections cover exactly what To200 needs — repo context, deploy logs, database metadata — and nothing else.

Policy gates

Autonomy settings decide when the AI may start work, push a branch, open a PR, or continue without review.

Balance-backed spend

Available credits bound total automation, while a per-task circuit breaker stops one failed route from looping indefinitely.

Evidence first

Every repair carries the original error, the affected code, the verification result, and a confidence score.

Audit trail

Project activity records what To200 saw, changed, verified, and handed to reviewers.

Model training

We do not use your Customer Content to train a public or shared model unless your organization separately agrees to that use.

What To200 reads — and what it never does.

Reads

  • Repository code and branches
  • Deploy and runtime logs
  • Error events and breadcrumbs
  • Database schema metadata

Never

  • Writes to main outside your policy
  • Broad, untracked account access
  • Secrets copied into prompts
  • Actions missing from the audit trail

FAQ

Questions, answered.

Can To200 merge directly to main?

By default, no: To200 stops at a pull request. Scale and High volume owners can deliberately enable merge-to-main, while repository rules and required reviews still apply.

What does To200 access in production?

Context through scoped integrations and monitoring events — deploy logs, error events, schema metadata. Never broad, untracked access.

What do reviewers see before approving a fix?

The original incident, affected files, the proposed change, the verification result, and a confidence score — one trail, one place.