At work I needed some clarity on how much each agent costs us and how they perform on production so I started looking at existing solutions for this such as LangSmith and LangFuse.
Honestly, they felt like they are overcomplicated to integrate and that there should be a simpler solution so I decided to build Tracium.
Tracium keeps track of every single AI API call that you make to an LLM provider, with this you can see traces, spans, I/O, tokens in/out, latency. You can compare during A/B testing, see usage from each customer and see which agents cost you the most vs which ones are used the most.
Feel free to check it out and sign up. This still is an early stage product so any feedback is welcome.
recognity•1h ago
Curious: do you track token costs per-agent or per-call? Being able to attribute costs to specific agent workflows would be a killer feature for teams running multiple agents in production.
antonijs•1h ago
Costs are calculated per call and then grouped per agent. So, the user can see how much each call within the agent costs as well as the entire agent.