Skip to content

Integration

GitHub Actions logo

GitHub Actions

Run Hyperpriors evaluation suites in CI and gate merges on model behaviour, exactly as you gate them on tests.

Request access to connect

Connections are provisioned in private beta — not authorised from this page.

01 — Permissions

When connected, Hyperpriors is limited to the following — declared up front, revocable at any time. Scope is granted during provisioning, not by a button on this page.

  • 01

    Trigger Hyperpriors evaluation suites from GitHub Actions workflows

  • 02

    Report eval results as status checks on pull requests

  • 03

    Block merges when evaluation scores regress below defined thresholds

  • 04

    Post summary reports of eval runs to pull request checks

02 — Details

Built by
GitHub
Website
github.com/features/actions (opens in new tab)
Category
CI/CD
Docs
docs.github.com/actions (opens in new tab)

03 — Notes

About GitHub Actions

GitHub Actions is GitHub’s integrated CI/CD platform. Workflows are defined in YAML, live alongside the code they build, and run on hosted or self-hosted runners in response to repository events — pushes, pull requests, releases, and schedules. Status checks from workflow runs can be marked as required, giving teams a merge gate that is enforced by the platform rather than by convention.

What the integration does

Hyperpriors turns evaluation into a CI step. A workflow step invokes your evaluation suites against the models, prompts, and agent configurations in the branch under review, then reports the results back to the pull request as a status check. Set thresholds on the scores you care about, mark the check as required, and a prompt change that quietly degrades behaviour can no longer merge quietly.

Results are summarised on the pull request itself: which suites ran, which cases regressed, and how scores moved against the baseline on your default branch. Full traces for every eval case remain available in Hyperpriors for the cases worth investigating.

This is the discipline the rest of your codebase already enjoys, extended to model behaviour. Prompts, model versions, and agent logic change constantly; the eval gate ensures those changes are measured before they are merged.

Get started

Add the Hyperpriors step to your workflow with an API key from your dashboard, or contact us for help designing your eval gates.