Tracing
Read an agent run like a stack trace
Multi-step agents fail in the middle. Cloudmind nests every retrieval, tool call and model invocation under one trace so you can see exactly which step went wrong.
Designed for systems with more than one model call
Nested spans
Arbitrary depth. A router calls a planner, which calls three tools, which each call a model. All one trace.
Retrieval visibility
See the query, the chunks returned and their similarity scores next to the completion they fed.
Tool call inspection
Arguments the model chose, what your function returned, and how long it took.
OpenTelemetry compatible
Export spans to your existing OTel collector, or ingest from one. No parallel observability stack.
Replay
Re-run any trace against a different prompt version or model and diff the two side by side.
Session grouping
Stitch multi-turn conversations into a session view so you can see how context accumulated.
Frequently asked questions
- Does this work with LangChain?
- Yes. The LangChain callback handler maps chains, agents, retrievers and tools onto Cloudmind spans automatically, so you get the full tree without manual instrumentation.
- What about frameworks you do not support?
- Use the generic span API. Any code you wrap in cloudmind.span() becomes part of the trace tree, which is how teams instrument custom agent loops.
Stop guessing whether your AI feature got better
Free for 50,000 traces a month. No credit card, no sales call, five minutes to your first trace.
Start free