Greenflash reads and analyzes every conversation between your agents and users and surfaces where user experience actually breaks down, even when eng metrics look nominal. So far we've run over 12 million analyses across 330,000 conversations.

My co-founder and I live this every day. When we first started dogfooding Greenflash, it exposed us to issues we never thought about during development. One quick example: early on in our agent's lifecycle, our inbox flagged a number of conversations where users were struggling. None of our other observability tools alerted us, and they couldn't, because nothing was technically broken. No errors, no failed calls. A few of our users were asking, and re-asking our agent for the same answers, becoming increasingly frustrated over the course of their conversations. Every dashboard said those conversations succeeded. What Greenflash saw was that the users never got what they came for. Its analyses surfaced the root cause on their own: our agent was truncating its outputs due to technical limitations, so the core of what people asked for never came back. We assumed it was a subset of our users and segmented to find them. Turns out there wasn't a boundary, it was our entire user base. From there we used our agent skill to pull the analyses and pair them with Claude Code, and shipped a fix to our core logic. The fix was easy. But understanding something was going wrong would be borderline impossible without Greenflash.

Users will do crazy things with your product, and test cases can't really capture UX failures. With language as the interface, this happens at a far greater rate. This is the space we care about most: the gap between what your users want out of your product, and the experience your agent actually provides them.

Under the hood, Greenflash analyzes every message, tool call, and conversation your users have with your product, and those analyses surface the patterns worth looking at out of the box. The agent skills are the part we lean on hardest: you point them at our analyses of your conversations to pull evidence fast, or to dig across trends that aren't obvious on the surface, what's moving the needle, what a proposed fix would actually do. It's all reachable through the API too, so you can pull an analysis or hand it straight to Claude.

Today, we support automated prompt optimization based off user conversations. We surface concrete prompt changes built on our analyses, delivered via API and webhooks. The caveat is that it takes upfront wiring to set up properly, and you should still keep a person in the loop as a validation gate. That's how we run it internally, and it hums along quite well. That said, prompts are only one part of the picture, as the story above shows.

Pricing is simple: free up to 1M tokens/month, then $299/month plus $5 per additional million tokens.

We're focusing on product owners and folks responsible for driving revenue-generating products. We're betting the person who owns the product, not the person who owns the infrastructure, is the one who should own agent quality. We're not sure teams are structured for that yet, or whether agent UX is still falling into a gap nobody owns. If you've figured out who owns this on your team, tell us how. Either way: sign up, try it, and poke holes.