OpenTelemetry

If your stack already emits OpenTelemetry traces, point them at Kelet. No re-instrumentation needed.

OTLP endpoint

https://api.kelet.ai/api/traces

Authentication

Authorization: <your-api-key>
X-Kelet-Project: <project-name>

Note: no Bearer prefix — just the raw API key.

Python

from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import BatchSpanProcessor
from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter

exporter = OTLPSpanExporter(
    endpoint="https://api.kelet.ai/api/traces",
    headers={
        "Authorization": "your-api-key",
        "X-Kelet-Project": "my-project",
    },
)

provider = TracerProvider()
provider.add_span_processor(BatchSpanProcessor(exporter))

Node.js

import { OTLPTraceExporter } from '@opentelemetry/exporter-trace-otlp-http';
import { BatchSpanProcessor } from '@opentelemetry/sdk-trace-base';

const exporter = new OTLPTraceExporter({
  url: 'https://api.kelet.ai/api/traces',
  headers: {
    Authorization: process.env.KELET_API_KEY,
    'X-Kelet-Project': 'my-project',
  },
});

provider.addSpanProcessor(new BatchSpanProcessor(exporter));

Multi-project routing

One process can emit traces to multiple Kelet projects. Tag a root span with kelet.project=<name> — all descendant spans inherit that project automatically.

with tracer.start_as_current_span("my-agent-root") as span:
    span.set_attribute("kelet.project", "my-project")
    # all spans below this go to "my-project"
    result = await agent.run(...)

Priority order per span (highest wins):

1. span.attributes["kelet.project"] — explicit subtree tag
2. Ancestor span’s resolved project — inherited automatically
3. resource["kelet.project"] — resource-level fallback
4. W3C baggage: kelet.project=... header — SDK propagation across services
5. X-Kelet-Project request header
6. "default" project

When to use direct OTEL

Use this path if:

• You already emit OTEL traces and want to add Kelet as a second destination
• You use a framework that generates OTEL natively (Langfuse SDK, LiteLLM, OpenInference-based frameworks)
• You need to integrate from a language without a Kelet SDK

Supported frameworks (via OTEL)

Tier 1 (tested, fully supported): pydantic-ai, Langfuse SDK, LiteLLM

Tier 2 (community-supported): Vercel AI SDK, OpenLLMetry/Traceloop, OpenInference/Phoenix

Tier 3 (via Tier 1/2 instrumentation): LangChain, LlamaIndex, CrewAI, Agno, OpenAI SDK, Anthropic SDK

Langfuse

Langfuse supports OTEL export natively. Point it at Kelet:

import os
os.environ["OTEL_EXPORTER_OTLP_ENDPOINT"] = "https://api.kelet.ai/api/traces"
os.environ["OTEL_EXPORTER_OTLP_HEADERS"] = "Authorization=your-api-key,X-Kelet-Project=my-project"

Or use the Langfuse integration for polling-based sync without any code changes.