Neon Docs

Agent ops: build, observe, evaluate, and improve.

Neon is an agent operations platform with durable execution, OpenTelemetry-native tracing, and evals-as-code SDKs. It helps teams ship reliable agents with clear, production-grade observability.

Start with Getting Started View repo

Quickstart path: Instrument an agent with OpenTelemetry, send traces to Neon, then layer in evals-as-code and regression checks as you scale.

What Neon Does

Observe any agent

Send OpenTelemetry traces from any runtime and query them fast in ClickHouse.

Durable execution

Temporal-backed workflows survive retries, timeouts, and approvals.

Evals as code

Use SDKs to define traces, scorers, and regression checks in code.

Two Integration Modes

Observe-only

Instrument your existing agents with OpenTelemetry and stream traces.

Managed execution

Run agents inside Temporal for durable, resumable workflows.

Quick Start

# 1) Add tracing (Python)
from opentelemetry import trace
tracer = trace.get_tracer("my-agent")

@tracer.start_as_current_span("agent-run")
async def run_agent(query: str):
    return await llm.generate(query)

# 2) Send traces to Neon
curl -X POST http://localhost:3000/api/v1/traces \\
  -H "Content-Type: application/json" \\
  -H "x-project-id: my-project" \\
  -d '{ \"resourceSpans\": [] }'

When to Use Neon

Online monitoring

Track live agent behavior and spot drift with traces.

Durable workflows

Keep long-running agent jobs resilient to failures.

Regression checks

Compare agent versions and catch quality drops before release.