Skip to content

Tbench Actions · built around your next push

Push code. Keep moving.

Run your CI on your own computer. Add Jira-linked AI repair, retest the change, and bring the result to your lead in Slack.

Approval first. Your 15-day trial starts after service approval. No automatic charge.

your-team / applicationWalkthrough
Terminal01 / 04

Push your branch

GitHub records the exact commit and workflow.

tbench / run
$ tbench-worker push
Push accepted · current branch
Waiting for its signed GitHub job event…
Step explained
Example workflow
  • Your existing workflowsGitHub jobs or an approved Tbench policy.
  • Explicit device consentA bounded session on your own computer.
  • Organization-led accessReviewed teams. Contributor seats.

01 / Two execution paths

Your pipeline. With a second pair of eyes.

Run your existing CI on your own computer, or add an approved AI profile for Jira-linked review, source-only repair and frozen retests. Your team lead keeps merge approval.

Explore the runner protocol
  1. GitHub workflow integration

    tbench-worker push

    Connect the GitHub App once, then keep your normal push and pull-request triggers. Run tbench-worker push from your checkout. Its bounded session handles the matching jobs on your computer.

  2. AI review, repair and retest

    tbench-worker push --ai-profile /path/to/review.json

    Select approved requirements and source files. The AI can propose a repair, rerun the unchanged tests and request lead approval through Slack. Models and repair limits come from your approved profile.

  3. Custom pipelines and named checks

    tbench verify

    Run named required or advisory application checks. Infinity’s separate 14-stage benchmark audit evaluates task quality, environment, reference solution and verifier.

02 / See the complete flow

From a failing push to a tested change.

Replay the recorded internal QA workflow: Jira requirements, an AI repair, unchanged tests, fresh CI and a real Slack approval.

Trigger
Jira-linked branch push invokes a bounded session on the developer computer.
Retest
Source-only repair runs against the unchanged repository test suite.
Approval
Team lead reviews the diff and test log in Slack before merge.
Tbench-AI / actions-qualificationRecorded QA

Execution trail 8 steps

Your computerBounded execution session
Terminal01 / 08

Push recorded

The original revision is retained, including its failure.

tbench / run
$ tbench-worker push
Original commit dd9b0a3b2c75
GitHub run 41341232214
Runner invoking computer
Recorded step passed
Recorded internal QA · 2 Nov 2026 · AI repair

03 / One organization. Clear control.

Your push. Your computer. Your workspace.

Each developer runs their own push on their own computer. Your organization manages repository access, contributor roles and the execution policy.

  1. 01

    One organization. One shared workspace.

    Apply with your work email. After approval, sign in with your Tbench account or configured company identity. Follow your first-run checklist and add approved teammates by email.

  2. 02

    Install once. Connect in your browser.

    Install the signed client and use your existing GitHub CLI identity. Organization and repository approval stay with your team. Run tbench-worker push from your checkout to start a bounded session on this computer.

  3. 03

    Your push starts your local session

    Keep your repository’s build and test steps. The command pushes the current branch, runs only its matching GitHub jobs on the invoking computer, then revokes the temporary session.

04 / Private by explicit policy

Your code deserves clear boundaries.

The independent runner reads approved local Git objects and keeps source and execution logs on the device. Tbench stores approved command policies and bounded job metadata. In the GitHub integration, GitHub holds your workflow logs and configured artifacts.

Remote AI services receive the inputs your workflow sends them. Define model hosting and allowed network destinations in your organization’s policy for source-sensitive review.

Explore execution and data controls

05 / Infinity Verifier & Experts

Author the benchmark. Audit the evidence.

Infinity’s 14-stage evaluation workflow audits the task, environment, reference solution and verifier. Issued results bind the outcome to the exact task and methodology revision.

06 / Start with a real workflow

Fifteen days. Your team. Your pipeline.

Apply with your organization’s work email. After approval, evaluate local Actions and application verification on your own compute.

  1. ApplyOrganization and repository access stay reviewed.
  2. RunYour enrolled devices and approved workload limits stay in your control.
  3. DecideNo automatic charge. Hosted compute and benchmark audits use separate approvals and budgets.
Try free for 15 days