The agent project library
Every agent project in the index, with what it burns and what it costs. Search by name, model or tag, filter by domain, then sort by burn, price or reliability — the source behind each number is one click away.
🔍
No projects match these filters
Try a shorter search term, a different domain, or reset the filters to see the full library.
How to read these numbers. Tokens per task is the measured context the agent consumes to finish one task — input, tool results and output. Cost per task applies the project's model blend to that token count. A success rate is the share of runs that end in an accepted result, not just a finished loop. Cost per solved task divides one by the other, which is the number that decides whether an agent is affordable at scale.
Provenance badges work the same way on every card: Verified means a public benchmark, paper or official doc states the metric, Community means operators reported it, Estimated means we modelled it from public signals.