Hosted Quickstart
Use this guide to send your first real AI trace to the hosted TraceLLM product.
1. Create A Project
Open the TraceLLM dashboard:
Create an account, open the dashboard, and copy a project API key from API Keys.
Full API keys are shown only once. If you lose one, create a new key and revoke the old key.
2. Install The Node SDK
You can also use pnpm or yarn:
3. Configure Your App
Add these values to your application environment:
TRACELLM_ENDPOINT=https://api.tracellm.in
TRACELLM_API_KEY=trllm_your_project_key
TRACELLM_SERVICE_NAME=my-ai-service
The API key connects traces to your project and loads your project capture policy.
4. Send A Trace
import { TraceLLM } from "@use-tracellm/sdk-node";
const tracellm = new TraceLLM({
endpoint: process.env.TRACELLM_ENDPOINT,
apiKey: process.env.TRACELLM_API_KEY,
serviceName: process.env.TRACELLM_SERVICE_NAME ?? "my-ai-service"
});
const session = await tracellm.startSession({
name: "Chat request",
userId: "user_123",
attributes: {
workflow: "chat",
provider: "openai",
model: "gpt-4.1-mini"
},
input: "Explain vector databases in one paragraph."
});
const span = await session.startSpan({
name: "openai.chat.complete",
kind: "llm",
attributes: {
provider: "openai",
model: "gpt-4.1-mini"
},
input: "Explain vector databases in one paragraph."
});
// Call your real LLM provider here.
const answer = "Vector databases store embeddings so apps can search by meaning.";
await span.end({
status: "ok",
usage: {
inputTokens: 12,
outputTokens: 18,
totalTokens: 30
},
output: answer
});
await session.end({
status: "ok",
output: answer
});
5. Inspect The Trace
Return to the TraceLLM dashboard and open Trace Explorer.
You should see:
- a session named
Chat request - an
llmspan namedopenai.chat.complete - duration and status
- token usage if token capture is enabled
- input/output content only if content capture is enabled
6. Configure Capture Policy
Open Project Config to choose what TraceLLM stores:
- sampling rate
- prompt and output capture
- input capture
- output capture
- token usage
- metadata
- errors
- email redaction
- API key redaction
- ignored span kinds
Configuration is project-level, so SDKs can pick up changes without redeploying your app.
7. Optional: Export To Your Observability Stack
Open Exports if you want to forward project traces to your own OTLP HTTP backend, such as SigNoz, Honeycomb, Grafana Tempo, Datadog, or an OpenTelemetry Collector.
TraceLLM remains the LLM workflow UI. External exports are for teams that also want the same trace data in their existing observability platform.