I'm building an app with OpenAI and Anthropic and need one place to inspect all traces

A good way to do this is to use a single observability layer that can ingest traces from both OpenAI and Anthropic, then let you inspect them in one UI.

Good options

1. OpenTelemetry + your own backend

   • Best if you want maximum control.
   • Instrument both SDKs to emit spans/events.
   • Send to a tracing backend like Jaeger, Grafana Tempo, Honeycomb, Datadog, or New Relic.
   • Pros: vendor-neutral, flexible.
   • Cons: more setup.

2. LangSmith

   • Popular for LLM app tracing.
   • Works well with multi-provider apps.
   • Good for inspecting prompts, outputs, tool calls, latency, and errors.

3. Helicone

   • Strong for LLM request logging and trace-style observability.
   • Supports multiple providers and gives a single dashboard.
   • Often easy to add via proxy or SDK config.

4. Arize Phoenix

   • Good for tracing and evaluation workflows.
   • Useful if you want deeper analysis beyond logs.

5. Datadog / Honeycomb / New Relic

   • If you already use one of these for production observability, centralizing traces there is often the simplest.

Recommended approach

If you want the most straightforward “one place” for LLM traces specifically, I’d start with:

• LangSmith if you’re building agent/workflow-heavy apps
• Helicone if you want quick unified request logging and trace inspection
• OpenTelemetry if you want a long-term vendor-neutral standard

What to capture

Make sure your tracing includes:

• provider (openai, anthropic)
• model name
• prompt/messages
• tool calls
• token usage
• latency
• errors/retries
• session/user/request ID
• parent/child spans for chained calls

Simple architecture

Your app → instrumentation layer → OpenTelemetry/LLM observability backend → single trace UI

If you want, I can also give you:

• a provider-agnostic tracing setup
• a comparison table of LangSmith vs Helicone vs OpenTelemetry
• or code examples for OpenAI + Anthropic in Node/Python.