Per-call cost tracking
Breakdowns per agent, per model, and per time period, with anomaly detection on cost per call.
Specialized · AI Tracing
See every AI call, its cost, and its conversation.
Monitor LLM costs, token usage, latency, conversations, and tool calls across every provider. Works with OpenRouter, OpenAI, Anthropic, and any OpenTelemetry-compatible provider.
Service facts
The meter on the machine that bills you
What ingest records for every AI call.
- 7profanity languages
- English, German, Spanish, French, Italian, Portuguese, Serbian — plus your own custom per-project terms.
- P95outlier line
- Anything above the 95th percentile on cost or conversation length is highlighted automatically.
- 0code changes on OpenRouter
- Enable Observability in your OpenRouter settings and point it at Traceway. That's it.
Metered honestly
Spot the spike before the invoice.
A single prompt with a large context window can cost 100× more than average; per-call visibility catches it the day it happens. Included in the open-source build and every Cloud plan — MIT, full parity.
- Cost
- Per call, per modelInput, output, and total cost per call; aggregated per agent and per model; cached and reasoning tokens tracked.
- Conversations
- Full chat timelinePrompts and completions rendered as chat, tool calls inline with arguments and paired results, sub-agents nested.
- Per user
- Stable user.idConversation count, median length, and cost per conversation per customer — never a session id or random value.
- Guardrails
- Scanned once, at ingestFlagged conversations get a badge and a filter, and can trigger alerts — no expensive query-time content search.
- Capture
- gen_ai.* attributesOpenRouter has built-in support; any OTel-instrumented AI call with gen_ai.* attributes is captured automatically.
On this line
Costs, conversations, and the outliers.
Calls group into conversations automatically; everything rolls up per user, per model, and per agent.
Conversation replay
See the exact prompt and the full response, tool calls with arguments and results, and a raw JSON view for edge cases.
Per-user analytics
How long conversations run, what they cost, and who your heaviest users are — with removable filter pills by user, model, or tool.
Latency at every percentile
P50 and P95 duration breakdowns per agent and model, throughput and token trends, providers compared side by side.
Privacy mode
Content is optional.
Prompts and completions are stored separately from trace metadata in object storage. Self-hosted, they never leave your infrastructure; OpenRouter also offers a Privacy Mode that sends only metadata — model, tokens, costs.
Next stop
Trace AI like any other service.
Costs, tokens, latency, and conversations, all in one place.