OpenTelemetry
Each AI call, except its content. Model, provider, tokens, service, environment and agent. Conversation, tool name and why the reply ended. From Claude Code and Codex: sessions, lines, commits, errors and limits hit. Never prompts, replies, instructions, commands or tool data.
Access
- Kind
- A key we issue; your side sends
- What to give
- An endpoint and a key we issue
- Why that one
- Your instrumentation sends to us. There is no key of yours to hand over and nothing for Valistry to call.
What it attributes on its own
- From the source
- request; model; provider; input and output tokens; service; environment; agent; conversation; prompt name and version; trace; latency; errors; Claude Code and Codex sessions, lines, commits, errors and limits hit; Cowork tokens, prompts counted and tool calls
- Needs a rule or a trace
- the bill: OpenTelemetry reports tokens, not money, so these rows use the vendor's list price and say so; a model with no list price, which shows at zero and is marked; anything your instrumentation does not emit
Connecting it
- Create the key. Your tools send this key with their OpenTelemetry data. It is shown once.
- Set up your tool. Pick the tool that makes your AI calls.
- Send, then check. Claude Code sends each minute. Still nothing? Your tool's log says why.
Also connecting a gateway or the provider for the same calls? Connect one, or each call counts twice.