Four Git platforms · fully automated PR review
Every pull request,
reviewed by an AI reviewer
before it merges
rs runs on your own servers. The moment a PR opens, it reads the entire change carefully and writes merge guidance and line-level comments straight onto the PR — a clear green light when it is safe to merge, and the exact line when something is wrong. No review rota, no waiting: one tireless gate in front of every merge.
- Reviews in 6–9 seconds
- Under ¥0.5 per review
- Chinese-language reviews available
✅ CAN MERGE 6 reviewer perspectives combined · 6–9 s
6–9s
From push to verdict
< ¥0.5
Cost per review
3 / 3
All 3 known issues in the demo PR caught
6
Built-in reviewer perspectives, mix and match
2.2×
Review throughput vs traditional CI review pipelines
0
Zero build-resource occupancy — a dedicated review service carries every review
Where the 2.2× comes from: reviewing on every push, multi-round merge gating, and reviews running in parallel — a single review takes no longer; a traditional CI review pipeline ties up build resources for ~10.6 minutes per review, and that time now goes back to your builds.
Platforms
A reviewer wherever your repos live
One review pipeline across all four platforms: set up a webhook once, reviews start the moment a PR opens, and merge guidance with line-level comments lands back on the original platform.
-
Gitea
Lightweight self-hosted Git — rs itself runs on Gitea, so it gets first-line support.
Supported -
GitHub
The de facto standard for open source and collaboration — one webhook and you are done.
Supported -
GitLab
Works with self-managed CE / EE and gitlab.com alike.
Supported -
Gitee
The go-to host for many teams in China — plug in on gitee.com and reviews start.
Supported
On another platform? Tell us what you are using: contact@sun-praise.com
01 How it works
How one review unfolds
The whole pipeline runs in the service you deploy. Work starts the instant a PR event arrives — no polling, no interruptions.
-
A PR event arrives
Set up a webhook on your Git platform once; every opened PR and new push notifies rs automatically.
-
The code is checked out, the changes located
rs fetches the PR's code and computes what changed; every later round focuses on the new deltas.
-
The AI reviews, file by file
Reviewers actually read the code and check evidence instead of skimming the diff; in team mode several perspectives run at once.
-
The verdict lands on the PR
Merge guidance and line-level comments appear directly on the PR; cost and elapsed time stay queryable in the dashboard throughout.
02 Features
Beyond review quality — easy to live with
Everything here is shipped and covered by tests — not a roadmap.
-
[01]
Team review perspectives
Six perspectives — quality, security, performance and more — review in parallel and fold into one verdict; drop a yaml file in the repo to add your own perspective.
qualitysecurityperformancearchitectureregression-testtest-value -
[02]
Automatic merge gating
Auto-merge fires only on a clear CAN MERGE with CI green; anything conditional or problematic is held for a human to decide.
✅ CAN MERGE auto-merges ⚠️ CONDITIONAL waits for a human 🚫 CANNOT is blocked -
[03]
Cheaper with every round
From the second push on, the reviewer remembers what it already said and focuses on the new changes — faster and cheaper.
-
[04]
Read-only, never executes
Reviewers can only read code — they never run commands on your machines. A stricter tier withholds shell access entirely.
-
[05]
Visible and controllable throughout
Misfired reviews can be stopped at any time and failed ones retried with one click; conflicting PRs skip outright, so no model spend is wasted.
-
[06]
Costs in plain sight
What each review cost and how long it took, summarized on one dashboard page; the same data ships over a read-only API and MCP endpoints.
-
[07]
Bring your own model
Any OpenAI-compatible endpoint plugs in, and traffic can split across models by ratio so you can compare quality and cost.
DeepSeekMiMovLLM… -
[08]
Lightweight self-hosting
One binary plus one Redis is a full deployment, data stays local, and setup takes minutes — no Kubernetes, no cloud dependency.
03 Review modes
One reviewer for everything,
or a whole team in parallel
Both modes work out of the box, globally or per repo — the light one for docs repos, the full team for critical business code.
Single reviewer
One AI reviewer reads the whole change — simple and direct, plenty for personal projects and small teams.
- Verdicts are plain: ready to merge / minor issues, mergeable / must-fix problems found
- Lowest cost — one review per push
- Later commits continue from the last round — no duplicated work
Team review
Several perspectives gate in parallel, then a lead reviewer synthesizes them — for repos with high quality bars.
- One clear verdict at the top: ✅ merge it / ⚠️ merge with conditions / 🚫 do not merge
- Each perspective folds away — expand any of them when you want the detail
- A failing leg is flagged plainly and the PR is held — never a disguised pass
04 Get started
Ready to give your team a review gate that never tires?
Private deployment and enterprise support available. Want to talk fit first? Send an email — we reply within one business day.
contact@sun-praise.com · Sun-Praise Technology Co., Ltd.
An empty API token only fails further down the line — better to fail fast right at the entry point. Suggested fix attached.