INTEGRATION / LLM PROVIDERS

Full visibility into every Claude conversation.

Whether Claude is drafting your support replies or running agentic workflows with tool use, Obsivara traces it all: every message, every tool call, every token — with cost, latency, and reliability rolled up per workflow.

ClaudeEXAMPLE
CALLS TRACED / 24H128,412
TOKENS ACCOUNTED / 24H41.2M
SPEND ATTRIBUTED / 24H$214.60
ILLUSTRATIVE · EXAMPLE DATA
WHAT YOU GET

Claude, fully observable.

Token and cost tracking per call

Input, output, and cache-read/write tokens for every request, priced accurately per Claude model tier and attributed to the calling workflow.

Tool-use and agent tracing

Claude's tool calls traced step by step — which tools were invoked, with what arguments, how long each took, and where agentic loops go off the rails.

Model tier economics

Compare Opus, Sonnet, and Haiku on your real workload. Find the workflows paying frontier prices for tasks a smaller tier handles just as well.

Cache efficiency insights

See your prompt-cache hit rates and what caching is saving you — and which workflows are leaving easy cache wins on the table.

HOW TO CONNECT

Three steps. No code changes.

1STEP 01

Connect your Anthropic account

Add a usage-scoped connection or route SDK traffic through Obsivara's proxy. Zero changes to your production prompts.

2STEP 02

Automatic workload mapping

Obsivara inventories every Claude model in use, the workflows behind each, and their baseline cost and latency profile.

3STEP 03

Turn on guardrails

Budget caps, latency SLOs, and anomaly alerts per workflow — configured once, enforced continuously.

WHAT WE MONITOR

Signals tracked out of the box.

  • Tokens per message (input / output / cache)
  • Cost per call and per workflow
  • Tool-use invocations and durations
  • Latency by model tier
  • Stop-reason and refusal patterns
  • Prompt-cache hit rate and savings
INTEGRATION FAQ

Common questions.

Does Obsivara support Claude's tool use and agentic patterns?

Yes — tool calls are first-class in Obsivara traces. You see each invocation, its arguments and duration, and the full loop structure of agentic runs.

Can Obsivara track prompt caching savings?

Yes. Cache reads and writes are tracked separately, so you see exactly what caching saves per workflow and where hit rates could improve.

Does this work with Claude on Bedrock or Vertex?

Yes. Claude usage through AWS Bedrock is covered by the Bedrock integration, and Obsivara consolidates it with direct Anthropic API usage in one view.

Connect Claude in minutes.

No credit card · 5-minute setup