PolyLever

● How It Works

From sign-in to AI review

Every push on a connected repository flows through the same automated pipeline — GitHub webhook, diff fetch, LLM review, and dashboard results.

01Step 01 of 07

Sign In

Create a PolyLever account with email, Google, or GitHub OAuth. Your session JWT unlocks the dashboard and GitHub App setup.

Authentication options

Three ways to sign in, one dashboard experience.

Email and password registration with immediate account access.
Google OAuth redirects through the backend and returns a session token.
GitHub OAuth links your GitHub identity for a faster onboarding path.

POST /auth/loginGoogle OAuthGitHub OAuth

02Step 02 of 07

Install GitHub App

Install the PolyLever GitHub App on your account or organization and select repositories. Each repo syncs as a project.

Repository connection

Install once, sync every selected repo as a project.

The install URL comes from the backend and opens GitHub’s app settings.
Installation webhooks register repositories under your PolyLever account.
Status endpoint shows which projects are connected and ready for pushes.

GitHub AppInstallation webhookProject sync

03Step 03 of 07

Push Code

Push commits to a connected branch. GitHub sends a push webhook with before_sha and after_sha to start the review pipeline.

Push detection

Every push on a connected repo starts a new review record.

GitHub delivers a signed push payload with branch and commit range.
PolyLever captures before_sha and after_sha for the diff window.
A review is created immediately with status pending or processing.

Push webhookCommit rangeBranch aware

04Step 04 of 07

Fetch Diff

The worker clones or updates the repo using a short-lived installation token, then runs git diff across the pushed commit range.

Diff preparation

The worker uses short-lived installation tokens — never stale PATs.

Clone or fetch the repository using a refreshed installation token.
Run git diff across the pushed commit range and store the patch.
Attach diff_content to the review before testing and analysis begin.

Installation tokengit diffWorker queue

05Step 05 of 07

Test Guardrails

100+ automated tests run as a safety check against faulty AI-generated reviews. The full suite is actively being built.

Test guardrails

100+ automated tests guard against faulty AI-generated output.

A growing suite of integration and regression tests validates the review pipeline.
Tests catch low-quality or incorrect AI output before it reaches your dashboard.
Full guardrail coverage is rolling out — the suite is actively being expanded.

100+ testsQuality gateComing soonRegression suite

06Step 06 of 07

AI Review

The diff is sent to the configured LLM. The model returns a markdown review covering bugs, security, performance, and maintainability.

LLM analysis

The model reviews bugs, security, performance, and maintainability.

Gemini 2.5 Flash is the default provider; OpenAI and Anthropic are configurable.
The prompt includes the full diff and asks for structured markdown feedback.
Empty responses are rejected so reviews never complete without content.

Gemini defaultMarkdown outputMulti-provider

07Step 07 of 07

View Results

Open the project review history in the dashboard. Each review shows status, the AI write-up, and the underlying diff side by side.

Dashboard results

Review history and detail views in one place.

List reviews per project with status badges and latest commit SHA.
Open any review to read the AI write-up next to the underlying diff.
Processing reviews show a loader until the worker finishes.

Review historyDiff viewerStatus tracking