Just launched LGTM: AI code review that doesn’t mark up your OpenAI tokens
Been lurking on IH for years, finally shipping something. Here’s the honest version.
I’m Tarin. Solo dev in Bangalore. Trade as DevsBazaar. Today I’m launching LGTM (Looks Good To Meow), an AI code review tool I’ve been building since early this year.
Why I built it: existing AI code review products fell into two buckets I didn’t like. Bucket one, they resell OpenAI tokens with a 3-4x markup baked into per-seat pricing. Bucket two, they run a single LLM prompt and confidently flag “security issues” that turn out to be false positives. After I got burned twice, I decided to build my own.
What LGTM actually does:
1. Every PR gets reviewed by 6 specialist agents running in parallel: security, bugs, performance, readability, best practices, docs. Each has a narrow system prompt. A synthesizer reads all 6 reports, resolves conflicts, dedupes overlap, and posts one comment with the final verdict.
2. A CI/CD scanner runs 16 deterministic rules against workflow files, Dockerfiles, and shell scripts. Things like pull_request_target abuse, hardcoded secrets, unpinned actions, privileged containers. Deterministic because I benchmarked LLMs on these patterns and they miss too many.
3. The Runtime Watchdog is a GitHub Action that fails the pipeline if the review verdict is “block”. This is how you gate bad merges without babysitting.
The BYOK part is the reason I’m even competitive. You bring your own OpenAI, Anthropic, or Gemini key. Your tokens show up on your provider bill, not mine. I don’t touch them. Yes this means users pay separately for LLM usage. That’s a real tradeoff, but I’d rather be honest about it than mark up 3x and pretend I’m doing them a favor.
Pricing (in INR because Bangalore, USD for reference): - Free: ₹0/mo, 10 reviews + 5 scans, 1 repo, no card - Hobby: ₹399/mo (~$5), 100 reviews + 50 scans, 3 repos, auto-review, Runtime Watchdog - Pro: ₹999/mo (~$12), 500 + 300, unlimited repos, custom detectors - Enterprise: custom, DPA/MSA, optional self-host
Top-ups for occasional spikes: +50 reviews (₹149), +250 (₹599), +100 scans (₹99), +500 (₹349).
Numbers so far: 2 months in beta, ~180 users signed up, 11 paid (all Hobby, no Pro yet), MRR ₹4,389. Not exciting but it’s real revenue from strangers, which I’ll take.
Stack: Node/TypeScript backend, MongoDB, React dashboard, GitHub App for repo install. Deployed on Fly.io. Payments via Dodo Payments (India Merchant of Record so I don’t have to handle GST myself). CLI: npm i -g @tarin/lgtm-cli .
What I need help with: - Feedback from anyone who reviews PRs on paid AI tools today. What do you hate about them? - Distribution ideas. My marketing budget is zero and I don’t have an audience. People willing to try the free tier and tell me what’s confusing.
Links: - Website: https://looksgoodtomeow.in
- Sign up: https://app.looksgoodtomeow.in
- Product Hunt: https://www.producthunt.com/products/lgtm-looks-good-to-meow?utm_source=other&utm_medium=social
- Bugs and feature requests: https://github.com/tarinagarwal/lgtm-feedback
- CLI: npm i -g @tarin/lgtm-cli
Ask me anything. I’ll be in the comments all day.
Congrats on the launch Tarin,
The BYOK model hits a sweet spot. For small teams and indie devs, the real question isn't just which tool reviews code, but whether it enforces secure coding practices without drowning everyone in setup.
A lightweight tool with a focused security ruleset often beats a heavyweight platform that nobody on the team actually uses consistently. Combining deterministic CI/CD checks with narrow LLM prompts is a smart move to keep false positives down while keeping token costs predictable.
Giving the CLI a spin on one of my side projects this week. Cheers from a fellow builder!
The BYOK decision stood out to me.
It doesn't just lower costs—it changes the trust relationship. Instead of asking developers to trust another AI vendor with both their code and their token spend, you're narrowing your responsibility to the review workflow itself. That's a very different position in the market, and I'd lean into that distinction more than competing on the number of agents.
Thanks for checking out LGTM! Happy to answer any questions about the architecture, multi-agent review pipeline, BYOK pricing, or the CI/CD security engine. I'd especially love feedback from anyone using AI code review tools today.