See inside every Haystack pipeline run.
Haystack wires retrievers, rankers, and generators into pipelines that answer from your data; Obsivara tells you what each answer costs and where it fails. Every pipeline run traced component by component — each retrieval, embedding, and LLM call carrying tokens, latency, and dollars — so a slow or expensive RAG pipeline points to the exact component. Instrument via OpenTelemetry or the Obsivara SDK; your pipeline runs unchanged, out of the request path.
Haystack, fully observable.
Component-level tracing
Every pipeline run traced through each component — retrievers, rankers, prompt builders, and generators — with inputs, outputs, latency, and the embedding and LLM calls each made.
RAG cost attribution
Embedding, retrieval, re-ranking, and generation priced and rolled up per query and pipeline, so you see which pipeline designs are expensive by construction.
Retrieval and latency visibility
How many documents each query retrieved and where time goes — retrieval versus generation — so a slow pipeline points to the exact component, not 'RAG is slow'.
Failure attribution and health
Component errors, retriever misses, and generation failures classified and traced to the component that broke, with health scoring and predictive alerts.
Three steps. No code changes.
Instrument with OpenTelemetry or the SDK
Point a Haystack OpenTelemetry tracing integration at Obsivara's OTLP endpoint, or wrap your app with the Obsivara SDK. Your pipeline keeps calling models directly; Obsivara records each run out-of-band, with no proxy in the path and no added latency.
Run your pipelines
Traces flow as components execute — no changes to your pipeline definitions, components, or document stores.
See it in Obsivara
Pipelines appear in the inventory and knowledge map with per-run cost, latency, and health within minutes.
Signals tracked out of the box.
- →Pipeline and component traces
- →Tokens and cost per query and pipeline
- →Embedding vs generation cost split
- →Retrieval latency and documents per query
- →Error rates and failing components
- →Health score and predictive alerts
Common questions.
Does Obsivara work with Haystack 2.x pipelines?
Yes. Haystack's tracing emits spans for each component in a pipeline; point its OpenTelemetry integration at Obsivara, or use the Obsivara SDK.
Can Obsivara track RAG cost per query?
Yes. Embedding, retrieval, re-ranking, and generation calls are priced and rolled up per query and pipeline, with waste detection on expensive retrieval patterns.
Do I need to change my Haystack pipelines?
No. You enable the OpenTelemetry tracing integration or the Obsivara SDK; your components, pipelines, and document stores stay exactly as they are.