Quickstart

This guide gets you from zero to your first trace, risk score, and cost attribution. You have two ways to start — use Splyntra Cloud (managed, nothing to run) or self-host the source-available core locally. Both speak the same OTLP API.

1. Get an endpoint and API key

Splyntra Cloud

1. Sign up at app.splyntra.com.
2. Create a project (or use the default one).
3. Copy your ingest key from Settings → API keys, or let the Connect an agent wizard mint one for you.

Your ingest endpoint is https://ingest.splyntra.com.

Self-host

Run the source-available core locally:

git clone https://github.com/splyntra/splyntra
cd splyntra
docker compose up

The collector listens on http://localhost:4318 (OTLP/HTTP) and the dashboard on http://localhost:3000. In development, the fallback key splyntra_dev_key is accepted (it is rejected in production). See Self-hosting.

note

API keys are stored only as SHA-256 hashes on the server — the raw key is shown once at creation. Keep it secret and pass it through an environment variable rather than committing it.

2. Instrument your agent

Splyntra is OpenTelemetry-native, so instrumentation is one import plus one init call. Set instrument to the frameworks and providers you use, and every step, model call, and tool invocation is captured automatically.

Python

pip install "splyntra[openai]"

import os
from splyntra import Splyntra

Splyntra(
    api_key=os.environ["SPLYNTRA_API_KEY"],
    project="support-agent",
    endpoint=os.environ.get("SPLYNTRA_ENDPOINT", "https://ingest.splyntra.com"),
    instrument=("openai",),   # add "langgraph", "crewai", "anthropic", …
)

# Run your agent exactly as before — spans are captured automatically.
from openai import OpenAI
client = OpenAI()
client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "What's our refund policy?"}],
)

TypeScript

npm install @splyntra/sdk

import { Splyntra } from "@splyntra/sdk";
import OpenAI from "openai";

new Splyntra({
  apiKey: process.env.SPLYNTRA_API_KEY!,
  project: "support-agent",
  endpoint: process.env.SPLYNTRA_ENDPOINT ?? "https://ingest.splyntra.com",
  instrument: ["openai"], // add "langgraph", "crewai", "anthropic", …
});

// Use the OpenAI SDK as usual — spans are captured automatically.
const client = new OpenAI();
await client.chat.completions.create({
  model: "gpt-4o",
  messages: [{ role: "user", content: "What's our refund policy?" }],
});

Raw OpenTelemetry

Already emitting OTel? Point your exporter at Splyntra's OTLP/HTTP endpoint — no SDK required:

export OTEL_EXPORTER_OTLP_ENDPOINT="https://ingest.splyntra.com"
export OTEL_EXPORTER_OTLP_HEADERS="Authorization=Bearer ${SPLYNTRA_API_KEY}"
export OTEL_SERVICE_NAME="support-agent"

Splyntra reads the OpenTelemetry GenAI semantic conventions, so gen_ai.* LLM spans, token counts, and model names are picked up automatically. See Ingest & OTLP.

Using a framework like LangGraph, CrewAI, or the OpenAI Agents SDK? Just add its name to instrument — see Framework integrations. Using a provider like Grok, Gemini, Groq, or OpenRouter through an OpenAI-compatible endpoint? See LLM providers.

3. See your first trace, risk score, and cost

Run your agent once, then open the dashboard:

• Traces — pick your latest run to see the full span tree with latency, tokens, and cost. See Traces.
• Security — each run gets a risk score; drill in to see which span triggered a secret, PII, or prompt-injection finding. See Security.
• Costs — the run's spend is attributed per model and rolled up into the project. See Costs.

A run summary looks like this:

{
  "run_id": "run_5f3a…",
  "project": "support-agent",
  "duration_ms": 4210,
  "cost_usd": 0.0132,
  "risk_score": 12,
  "findings": []
}

Next steps

• How it works — the architecture and data flow.
• Manual instrumentation — trace your own agent, tool, and LLM functions.
• Guardrails — block or redact risky calls inline with guard.
• Evaluation — score outputs and gate regressions in CI.