edenObserve · Evaluate · Improve
See every agent run.
Fix what fails.
Drop-in observability for AI agents — traces, evals, and reviewable improvements in one loop.
- Overhead
- ≤5ms
- Ingest
- OTel
- Loop
- Closed
Observe
Traces that survive production
Sessions, spans, detectors, and issues — grouped so you can triage failures without grepping logs.
- SessionsGroup runs by user, agent, or deploy12 live
- DetectorsDrift, latency, and cost thresholds3 firing
- IssuesDeduped failures with first/last seenISS-1842
Control
0.84
Treatment
0.91
Live A/B · n=2,400 · pinned env: prod · Δ +0.07
Evaluate
Prompts, datasets, live A/B
Version prompts, pin environments, and compare arms on the metrics that matter before you ship.
Improve
Close the loop from real failures
Plans from traces you already have. Approve, reject, or snooze — then watch the next deploy with the same instruments.
- 01
Detect
Issue opens from a detector or manual triage.
- 02
Propose
Plan cites the failing spans and eval delta.
- 03
Ship
Human review — then re-observe the next run.
Install
Instrument in a few lines
Install eden-sdk or point traffic at the OpenAI-compatible gateway.
import eden
eden.configure(
org_id="org_…",
api_key="sk_eden_…",
)
eden.instrument() # OpenAI / Anthropic / LangChain / …Start
Instrument today. Improve tomorrow.
Free during public beta — no credit card.