A drop-in proxy between your apps and LLM providers. Full visibility into costs, tool calls, and agent behavior. Zero code changes — just change the base URL.
Everything QA and dev teams need to understand AI agent behavior
Spending breakdown by provider, model, and session. Know exactly where your AI budget goes — before it spirals.
Automatically correlates related API calls into logical sessions using conversation fingerprinting and tool-call linking.
Visualize complete agentic workflows: user prompt, LLM response, tool call, tool result. Debug broken chains instantly.
Automated PASS / REVIEW / FAIL checks per session. QA engineers see health status first, details second.
Each session rendered as a step-by-step narrative timeline. Readable conversations, not JSON blobs.
Generate clean bug reports from any session. Copy as plain text for tickets, or export raw JSON for developers.
One config change. Full observability.
Change your LLM SDK's base URL to the Inspector proxy. No code changes, no SDK wrappers.
# OpenAI Python SDK
client = OpenAI(
base_url="http://inspector:8080"
)
Every request and response is traced and stored. Streaming supported with zero added latency.
App → Inspector → OpenAI
↓
traces saved
Sessions, verdicts, conversation flows, costs — all visible immediately. Zero configuration.
# Dashboard ready at
http://inspector:8080
Sessions: 47
Total cost: $12.84
AI agent traces are opaque. We make them readable.
Built for everyone who needs to understand what AI agents are doing
Verify agent behavior, spot failures, generate bug reports. No developer skills needed — the story view speaks plain language.
Debug agent execution flows, trace tool interactions, inspect raw payloads. Full request/response detail for every API call.
Track LLM costs, performance, and reliability across sessions. Know what your AI spend buys — and where to optimize.
Drop-in proxy. Full visibility. Zero code changes.