// Show HN: PullLight

Code review you can
stop thinking about
without the noise.

Every AI comment is gated through human approval before it posts to GitHub. Unlike CodeRabbit, Greptile, and Copilot — nothing auto-publishes. Free for OSS. $20/mo flat for teams.

PullLight AI code review dashboard showing bug findings
0+ Total catches
0+ Critical bugs caught
25 Famous OSS repos scanned
$20/mo Flat team pricing
// real bugs, real repos

Not synthetic examples — actual catches.

These are bugs PullLight found in production open-source repos. CVSS scores from NVD.

● CRITICAL
Next.js middleware auth bypass
vercel/next.js
Middleware matched only exact paths, allowing attackers to bypass auth by appending URL fragments. Unauthenticated access to protected routes.
CVSS 9.1 · CVE-2025-29927
Read case study →
● CRITICAL
jsPDF path traversal
parallax/jsPDF
File-save handler did not sanitize the output path. Arbitrary write to any filesystem location the process had permission to access.
CVSS 9.2 · CVE-2025-68428
Read case study →
● HIGH
ip package SSRF bypass
indutny/node-ip
IPv6-mapped IPv4 addresses bypassed private-IP checks. Attackers could reach internal services from external requests.
CVSS 8.1 · CVE-2024-29415
Read case study →
See PullLight catch bugs in 25 famous OSS repos — /showcase →
// why this isn't more AI noise

This is the section that matters to HN.

Every other tool in this space trains engineers to ignore it. Here's why PullLight is different.

🔒

Every AI comment is approved by a human before it posts.

When the webhook fires, Claude analyzes the diff and queues findings at /reviews. Nothing auto-publishes. Zero. A comment that appears on a PR is one a human decided was worth the developer's attention.

🎯

Severity floor configurable per repo. We never flag style.

You configure the minimum severity per repo — show me CRITICAL + HIGH only, or everything. PullLight's prompt is tuned to ignore formatting, whitespace, and style nitpicks unless you explicitly ask for them.

💸

Free for OSS. $20/mo flat for teams — not per seat.

CodeRabbit charges $19–$24/dev/mo. Greptile $30/dev/mo + overages. At 10 developers, that's $190–$400/mo. PullLight is $20 for the whole team. The price is locked for early-access teams.

Feature ◈ PullLight CodeRabbit Greptile Copilot
Human approval gate Yes No No No
Pricing model $20 flat/team Per seat Per seat + overages Per seat
Noise level Low High Medium High
Free for OSS Yes Limited No With Copilot

Pricing verified from public sources June 2026. Full comparison →

// faq

Common questions.

How is this different from CodeRabbit?

CodeRabbit auto-publishes every comment it generates — directly to GitHub, on every PR, before any human sees it. PullLight queues findings for human review at /reviews. Nothing posts until you approve it. That's the core distinction. CodeRabbit is also $19–$24/dev/mo; PullLight is $20/mo for the whole team.

What does Claude actually see? Do you train on my code?

Claude sees the PR diff — the same text that appears in a GitHub PR's "Files changed" tab. It does not see your full codebase, your git history, or any private context beyond the diff. We do not train on your code, store diffs after analysis, or use your findings to improve models. Full data and security policy at /trust →

How do I uninstall?

Go to github.com/settings/installations, find PullLight, click Configure → Uninstall. That's it. No data is retained after uninstall. PullLight stops receiving webhooks immediately.

What languages does PullLight support?

Any language Claude can read — which is essentially every mainstream language: JavaScript, TypeScript, Python, Go, Rust, Java, C, C++, C#, Ruby, PHP, Swift, Kotlin, and more. PullLight reviews diffs, so it works on any file type Claude can analyze, including config files, SQL, and infrastructure-as-code.

Can I self-host PullLight?

Self-hosting is on the roadmap. The current version is cloud-hosted only — you install the GitHub App and reviews flow through PullLight's infrastructure. If self-hosting is a hard requirement for your team, see our data handling policy and reach out via the feedback button.