Developer tools
Instrumentation for the agent behind the user-facing app. Nothing here is visible at the front door.
Ask + trace
One question, one architecture, every step — the model calls, the tool calls, the tokens and the cost of each.
Compare
The same question through several architectures side by side, with the cheapest, fastest and best-cited arm called out.
Retrieval
Semantic, lexical and hybrid search without the agent. Use it to tell a retrieval problem apart from a generation problem.
Evaluate
The graded suite: recall@k, precision@k, MRR and nDCG for retrieval; route accuracy and answer matching for the arms.
Runs
Every run ever recorded, and aggregate cost and latency by architecture.