A SonarQube alternative for AI-written code
Catch the second copy before it merges.
Coding agents write a new helper instead of finding the old one. IonWarp flags the duplicate, the second code path and the code left behind.
9:41
‹PR #733Finding 1 of 3
AI ArchitectureP1
A second formatPrice drifts from the first
components/cart/price.ts
1+export function formatPrice(cents: number) {2+return '$' + (cents / 100).toFixed(2);3+}
lib/money.ts already formats prices with the user’s currency. This copy hard-codes dollars, so EU carts show the wrong symbol.
@claudeCopy prompt
Also on this PR
- P2v1 handler is still routedapp/api/v1/orders.ts
- P2Feature flag checked in three placeslib/flags/use-flag.ts
HomeBugsTeamMy Standup
Read-only code accessNever pushes a commit
How it works
Install the GitHub App
Pick the repositories. IonWarp reads the code and comments on pull requests. It never pushes a commit.
Open a pull request
AI Architecture reads the diff. Flags duplicate code, second code paths and code left behind.
Fix it with your agent
Every finding has a fix prompt for Claude Code, Cursor or Codex. Push the fix and IonWarp checks it again.
Specialist reviewers
Add Log & Errors
to your pull requests.
IonWarp is a next-gen AI code reviewer for GitHub pull requests, built on open models.
IonWarp Log & Errors vs AI reviewers:38% less wait · 69% cheaperon DeepSeek V4.1 Flash, same 17 PRs →
IonWarp Review — PR #958
| # Sev | Issue | File |
|---|---|---|
| 1 | Write failure swallowed by an empty catch | lib/orders/save.ts |
| 2 | Webhook promise rejection is unhandled | app/api/stripe/route.ts |
| 3 | Error logged without a request id | lib/log.ts |
27try { await db.insert(orders).values(row); }28+catch {}29return { ok: true };
The empty catch hides the insert error and the function returns ok. The customer sees a confirmation for an order that was never saved.
What AI Architecture catches
The second copy
A helper, type or query that already exists somewhere else.
Code left behind
Old routes, flags and fallbacks the change made dead.
One idea, many places
The same rule written in three files, ready to disagree.
How each reviewer scores.
On 11 real pull requests with 6 known issues, AI Architecture runs on DeepSeek V4.1 Flash: F1 47.1%, 6m 14s a review at the median, $0.050 in model cost. CodeRabbit leads on F1 at 72.7%, at ~$1.50 a review. This lane is provisional until its case set grows and its labels are reviewed.
| Model / App | Reviewer Score | F1 | Latency | $/run |
|---|---|---|---|---|
| CodeRabbit | 60.6 | 72.7% | 7m 43s | ~$1.50 |
| IonWarp · GLM 5.3 | 59.3 | 47.1% | 1m 30s | $0.075 |
| IonWarp · GPT 5.6 Luna Pro | 53.8 | 35.3% | 51s | $0.046 |
| IonWarp · DeepSeek V4 Pro | 53.4 | 44.4% | 1m 52s | $0.102 |
| Greptile v4.1 | 50.5 | 60.0% | 5m 06s | ~$1.75 |
| IonWarp · DeepSeek V4.1 FlashRuns on your PRs | 47.8 | 47.1% | 6m 14s | $0.050 |
| IonWarp · GLM 5.3 Flash | 38.3 | 15.4% | 1m 02s | $0.0064 |
| Cursor Bugbot | 37.4 | 50.0% | 7m 46s | ~$1.25 |
| IonWarp · Grok 4.6 | 34.0 | 34.8% | 3m 29s | $0.284 |
| IonWarp · Muse Spark 1.3 | 33.6 | 28.6% | 2m 04s | $0.193 |
| IonWarp · Gemini 3.8 Flash | 32.3 | 33.3% | 3m 50s | $0.238 |
| Qodo v2 | 12.8 | 25.0% | 5m 52s | ~$1.67 |
Reviewer Score = (3 × F1 + time + $/run) ÷ 5, each part 0 to 100 within this table. IonWarp rows cost the model’s inference; ~$ is each app’s list price for 5 seats at 100 reviews a month. Provisional scoped-label matching. Release 2026-10-12, CC BY 4.0.
AI Architecture
IonWarp inferenceApp price estimatesF1 on the same 11 open-source pull requests for every reviewer
Pricing
Simple pricing for your whole team.
One workspace price with seats included. Every agent on every plan, and a monthly budget IonWarp stays inside. 14 days free, no card.
ProMost popular
One team, every agent from day one.
- 5 seats included
- All eight agents, every repo you connect
- $25 of review spend a month
- A monthly budget you set
Max
Bigger teams and a bigger review budget.
- 10 seats included
- All eight agents, every repo you connect
- $100 of review spend a month
- A monthly budget you set
- Priority models and support
More than 10 people?
Tell us how many seats and how much review spend you need, and we will set up the workspace.
FAQ
Frequently asked questions
What is AI slop in code?
Code that works but repeats what exists, leaves old paths running, or adds a layer nobody needed. It is cheap to write and expensive to keep.
Is it a SonarQube replacement?
For duplicate logic and dead paths in a pull request, yes. SonarQube also tracks coverage and rules across the whole codebase; IonWarp reads the change and the code it touches.
Will it nag on style?
No. It only flags duplicate logic, dead paths and repeated rules, never formatting.
How accurate is AI Architecture?
On 11 real pull requests with 6 known issues, AI Architecture runs on DeepSeek V4.1 Flash: F1 47.1%, 6m 14s a review at the median, $0.050 in model cost. CodeRabbit leads on F1 at 72.7%, at ~$1.50 a review. This lane is provisional: its case set is small and still under review, so read it as a direction, not a ranking. Every row, PR and judge verdict is public.
What does it cost?
Pro is $49 a month per workspace with 5 seats and $25 of review spend included. Max is $149 with 10 seats and $100. Every agent is in both. Start with a 14-day trial, no card.
More on ai architecture
More on ai architecture

Let agents write. Keep one copy.
Try for free14 days free, no card. Then Pro is $49 a month per workspace with 5 seats, or Max $149 with 10.




