ChatGPT Subscription
- Account
- alex@example.com
- Plan
- Pro plan
Find bugs before you merge.
Tough Crowd reviews your GitHub pull requests and reports the bugs that need fixing. You choose Claude or Codex as the reviewer. Your team keeps its existing workflow, and Tough Crowd can run alongside the bug bot you already use.
Some checks were not successful1 failing, 3 successful checks
- Tough Crowd / code-review Failing after 1m — Not safe to mergeDetails
- CI / test (pull_request) Successful in 4mDetails
- CI / lint (pull_request) Successful in 48sDetails
- Your existing review bot Successful in 2mDetails
No conflicts with base branchMerging can be performed automatically.
Save money
Bring your own subscription or key.
Most bug bots charge you directly for inference. Instead, just connect your Claude or ChatGPT subscription, or add an Anthropic or OpenAI API key, and pay your provider directly. You only pay us for the time your sessions use.
Anthropic API key
...3kT2Encrypted at rest
Stay on the frontier
Choose the best agent, model, and effort.
Most bug bots select the model for you. You can decide which frontier model to run reviews with.
When Anthropic or OpenAI ship a stronger model, you can start to use it right away.
Also leverage the amazing work that goes into Claude and Codex by using their agents to conduct reviews.
Claude Code · Opus 5.5
agent
- Claude Code
- Codex
model
- Opus 5.5
- Sonnet 5.5
- Fable 5.1
- Fable 5
- Opus 5
- Sonnet 5
- Opus 4.8
- Sonnet 4.5
Medium
Reasoning effort
- LowFastest. Minimal reasoning.
- MediumBalanced speed and depth.
- HighDeep reasoning for complex tasks.
- Extra HighExtended reasoning for demanding tasks.
- MaxMaximum available reasoning depth.
Sharpen your reviews
Run multiple reviews simultaneously.
Run targeted reviews with specialized prompts for things like security.
Each review reports its own GitHub check, so branch protection can require the ones that matter.
Test reviews with a new model or new instructions next to your current setup and compare the results side-by-side on real pull requests.
.toughcrowd/config.yaml
version: 1 reviews: correctness: agentProfile: codex-cli-chatgpt model: gpt-6.1-sol instructions: ../reviews/correctness.md security: agentProfile: claude-code-cli-default model: claude-opus-5-5 instructions: ../reviews/security.md
Checks on this pull request
correctnessCodex · GPT-6.1 Sol1 blocking finding
securityClaude Code · Opus 5.5No issues found
Stay focused
Fix bugs without dropping what you're doing.
Spot a problem in the review, start a fix right from the comment, and get back to what you were doing. A coding session picks up every open finding.
It keeps going until the review passes. Tough Crowd re-reviews each push, and the session fixes whatever's left, back and forth until it's done.
You come back to a pull request that's safe to merge.
Tough Crowd
CautionNot safe to merge — correctness found 1 blocker.
correctness · Not safe to merge
Reviewed 9e8d7c6 2026-10-13 16:06 UTC · View session
Blocking (1)
-
Members can invite organization admins
src/api/invitations.ts:48· open since run 1 (9e8d7c6) The invite endpoint checks membership but does not require the inviter to be an org admin. A member can issue an admin invitation. To unblock: Require org admin permission before creating any invitation.Full finding
The invite endpoint checks membership but does not require the inviter to be an org admin. A member can issue an admin invitation.
Run history (1 run)
| Severity | Raised | Resolved | Open |
|---|---|---|---|
| Blocking | 1 | 0 | 1 |
| Warning | 0 | 0 | 0 |
| Notice | 0 | 0 | 0 |
Converging. 0 of 1 findings are resolved. The remaining blocker has been open since run 1.
| Run | Head | Reviewed (UTC) | Blocking | Warning | New | Resolved |
|---|---|---|---|---|---|---|
| 1 | 9e8d7c6 | 2026-10-13 16:06 | 1 | 0 | 1 | 0 |
One coding session fixes the open review findings and pushes once.
Updated 2026-10-13 16:06 UTC
Put your next pull request in front of a tough crowd.
Tough Crowd is free during early access.