Connect Obsivara to your AI
Obsivara captures what your AI agents actually do — runs, spans, LLM calls, token counts, model, cost, errors, and health — so you can monitor, debug, and improve them over time.
What Obsivara captures
Every time an agent runs, Obsivara records the full lifecycle: when it started, what inputs it received, every LLM call (model, prompt/completion tokens, cost, latency), every tool invocation, errors, and the final outcome. The data flows into a per-asset health view with trend charts, error rates, and cost breakdowns.
You do not need to change your agent's logic. You instrument from the outside — by sending a few HTTP calls at the start, during, and end of each run.
Three ways to connect
Pick the approach that fits your stack:
Generic webhook
Any language, any framework. One HTTP POST per event — no install required. The fastest way to get data into Obsivara.
Webhook guide →OpenTelemetry (OTLP)
Already using OpenTelemetry? Point your exporter at https://api.obsivara.com/ingest/otlp. Obsivara maps gen_ai.* / llm.* attributes automatically.
Let an AI agent wire it up
Hand an AI agent an API key and the connect guide. It can instrument your workflow end-to-end without manual steps.
Agent guide →New here?
The fastest path to your first event is the Quickstart — a single curl command that sends a run event and returns a 202 in under five minutes.