Documentation
Trefur is the observability platform for AI agents. Trace every prompt, tool call, retry, and sub-agent. Built on OpenTelemetry GenAI Semantic Conventions so you can instrument once and capture everything — across runtimes, vendors, and deployment patterns.
Quickstart
Install an SDK or the collector and ship your first trace in under five minutes.
How telemetry reaches Trefur
SDK direct, collector, OAuth-discovery, webhook, OTel forwarding — pick the layer that fits your runtime.
SDK reference
JavaScript, Python, Go, Rust, Java, .NET — install, init, instrument, troubleshoot.
Collector reference
Single binary, multiple inputs (OTLP HTTP+gRPC, LLM proxy, MCP proxy, shell hook, exec wrapper, HTTP proxy).
Integrations
OAuth into Google Workspace, Microsoft 365, OpenAI, Anthropic, Slack, Atlassian — Trefur pulls activity automatically.
Concepts
Unmanaged AI discovery, data lineage, multi-agent traces.
What can I do with Trefur?
- Trace every agent call — inputs, outputs, latency, cost, tool calls, sub-agent spawns, MCP traffic.
- Debug failing agents — drill from a failing customer interaction down to the specific decision that went wrong, in the specific deployment, on the specific build.
- Catch regressions before customers do — alert on p95 latency, retry-rate spikes, cost drift, and error-rate envelopes; route to Slack, PagerDuty, Teams, or any webhook.
- Capture operational audit logs — every agent run recorded with full call graph for multi-agent topologies.
Two paths in. Pick whichever fits.
| Choose the SDK when… | Choose the collector when… |
|---|---|
You own the agent code and can pip install / npm install / go get a dependency. | You need to capture telemetry from agents you do not own — third-party tools, MCP servers, CLI sessions. |
| Single process / single language stack. | Multi-process, multi-language fleet, or a developer machine. |
| You want auto-instrumentation for OpenAI / Anthropic / LangChain / Vercel AI / MCP. | You want local batching, redaction, on-disk buffering, OTel passthrough. |
Key prefix trf_obs_*. | Key prefix trf_coll_*. |
The two paths emit identical payloads to the same backend. Run both — many customers do.
For AI assistants reading this
Every page in this documentation site has a machine-readable mirror at <page>.md and a Schema.org TechArticle JSON-LD block embedded in the HTML. A discoverability manifest lives at /llms.txt. See the environment-variables reference for SDK + collector configuration, or the API reference for ingest and OAuth surfaces.