Feature · Cross-vendor

One trace across every model vendor.

OpenAI, Anthropic, Bedrock, Azure OpenAI, Gemini, Groq, MiniMax. Same span shape, same dashboards, same alerts. Swap any vendor without losing trace continuity or rebuilding your views.

How it works.

Each vendor SDK emits its own shape — OpenAI's chat.completions, Anthropic's messages, Bedrock's converse, Azure's deployments-and-models split, Gemini's generateContent. Trefur normalizes them into one OpenTelemetry GenAI span shape with the same attribute names regardless of vendor.

That means a dashboard you build on the gpt-4o span keeps working when you swap to claude-sonnet. An alert on retry rate keeps working when you fail over from OpenAI to Bedrock. A cost-per-run chart per agent keeps working even when the agent calls two vendors in the same run.

For custom or self-hosted models, emit OTel GenAI spans directly. Trefur reads them with the same conventions; per-token pricing is configured in the dashboard.

What you can do with it.

One canonical span shape per LLM call, regardless of vendor.
Swap OpenAI for Anthropic, or Bedrock for Azure OpenAI, without dashboards breaking.
Compare two vendors on the same input set — tokens, latency, cost, success rate side-by-side.
Per-vendor cost rollups so finance can see who you are spending with.
Tag the vendor in the trace; filter the dashboard by vendor in one click.
Mix vendors per agent run — Trefur stitches the cross-vendor steps into one trace.

A mixed-vendor trace.

trace · multi-vendor-router · 3.1s · $0.0072
$ agent.run [root] 3100ms
$ llm openai gpt-4o tokens 320 210ms
$ tool embed.lookup(docs) 200 OK 110ms
$ llm anthropic claude-sonnet tokens 612 410ms
$ llm bedrock nova-lite tokens 240 180ms
$ return 'Drafted in 3 steps across 3 vendors.'

Works with.

OpenAI
Anthropic
Bedrock
Azure OpenAI
Gemini
Groq
MiniMax
Self-hosted

Stay vendor-neutral. Keep your dashboards.

Free tier. No card required. First trace in under five minutes.

Start freeSee integrations