three.dev
Find recurring failures in AI agent conversations and compare behavior across releases.
Select a screenshot to enlarge it.
About this product
three.dev helps teams see how an AI agent behaves after deployment. It routes model calls through its proxy, scores a configurable sample, and groups problems into recurring failure modes. Each issue points back to the conversation evidence. A coding agent can retrieve that context through MCP, and the team can compare two versions using the same evaluation criteria.
Start with synthetic or redacted test traffic and define one failure you can judge by hand. For a support agent, that could be quoting the wrong return window. Compare failures as a share of scored requests, especially when comparing different traffic volumes or sampling rates. Check what conversation data the proxy records before connecting production traffic.
The site offers early access. Its pricing page lists 5,000 scored requests and 1 GB of processed data per month on the free plan, with Starter listed at $79 per month.
Features
- Conversation scoring and issue evidence
- Grouped failure modes
- Coding agent access through MCP
- Before and after traffic comparisons
Launches and product updates
Product announcements submitted for editorial review before publication.
No launch updates have been published here yet.
Reviews
Write a reviewExperiences from people who have used the product. Reviews are moderated before publication.
Share what worked for you
No community reviews yet. Your experience can help someone choose the right tool.
Discussion
Start a discussionAsk a question, share a workflow or talk with the product team. Messages are moderated before they appear.
No discussions yet. What would you like to know about this product?
Join the conversation
Explore similar tools
Other products in these categories. Compare their features and pricing for your task.
Product team
The product team has not claimed this listing yet.
Claim this product
Are you the maker or an authorized team member? Request access to manage this listing. Ownership is checked before access is granted. If approved, your account display name will appear publicly as the verified listing manager. Your verification evidence stays private.
Add a badge to your website
Link visitors to this profile with a live MustHave badge.
Add a MustHave badge
Show your MustHave listing on your website. The badge displays its current community vote count.