Skip to main content

Eden

Observability and improvement for AI agents.

Drop-in tracing, auto-instrumentation, deterministic judges, cost intelligence — and a portal that turns traces into reviewable patches.

QuickstartBrowse SDKsStart free
pip install eden-sdk

What is Eden?​

Eden is the data plane for production AI agents. Point an SDK or gateway at your traffic, Eden captures every LLM call (tokens, cost, latency, model, prompt, response), and the portal turns the firehose into traces, judge scores, cost breakdowns, regression alerts, and human-reviewed patch suggestions.

The platform ships:

  • Two production SDKs — Python (eden-sdk) and TypeScript (@eden-ai/sdk). Fire-and-forget; producer p99 overhead budgeted ≤ 5 ms.
  • Open ingestion — OpenAI, Anthropic, Bedrock, Gemini, LangChain, LlamaIndex, CrewAI, AutoGen, Vercel AI, and raw custom events. Wire it from your own code even without an SDK.
  • One portal — Next.js workspace at edenobservability.com. Traces, evaluators, cost intelligence, and Improve.

Observe. Evaluate. Improve.​

Observe

Capture every call

OTel-compatible ingestion, full trace timelines, and specialised intelligence surfaces.

Ingestion reference →

Evaluate

Score what shipped

Deterministic + LLM judges, regression alerts, golden datasets, and batch replay.

Judge tutorial →

Improve

Close the loop

Portal turns failing traces into reviewable patches — with budgets, PII redaction, and RBAC underneath.

Set a budget →

Pick a starting point​

New here

Integrate

Security

SDKs talk to one API​

Both SDKs hit the same ingestion surface (POST /orgs/{org_id}/ingest/{format}) at https://api.edenobservability.com.