Integrations
Open core · self-host + all Splyntra Cloud plans
Splyntra is OpenTelemetry-native. Everything you send — from an in-process SDK, a raw OTel exporter, or an out-of-process workflow platform — lands on the same OTLP pipeline, is redacted, scored for risk, and cost-attributed the same way. There is no proprietary wire format to learn.
The gen_ai.* semantic conventions
The collector reads the OpenTelemetry GenAI semantic conventions
(and OpenLLMetry's traceloop.entity.*) alongside Splyntra's own splyntra.*
attributes. That means model names, token counts, and inputs/outputs are picked up
automatically from any conforming source — third-party OTel GenAI instrumentation and
the hyperscaler agent platforms included — with no code changes on your side.
Three ways in
| Path | Best for | How |
|---|---|---|
| SDK auto-instrument | Frameworks and providers running inside your Python or TypeScript process | One line: instrument=(…). See Framework integrations and LLM providers. |
| Raw OpenTelemetry | Any language already emitting OTel GenAI spans | Point your OTLP exporter at the collector. See Ingest & OTLP. |
| Platform webhooks | Hosted / no-code workflow platforms that run out-of-process | POST a run summary to ${ingest}/v1/integrations/<platform>. See Agent platforms. |
All three speak Authorization: Bearer <ingest-key>, and all three produce the same
trace + risk + cost data model.
Support tiers
Integrations sit in one of four tiers:
| Tier | Meaning | Examples |
|---|---|---|
| Native OTel | Emits gen_ai.* spans that ingest directly at /v1/traces, no glue | Any OTel GenAI source; AWS Bedrock AgentCore, Google Vertex Agent Engine |
| Auto-instrument | One-line SDK adapter captures the framework in-process | LangGraph, CrewAI, OpenAI Agents, LlamaIndex, MCP, Browser Use |
| Cost / provider | Cost-tracked and labeled by provider via an OpenAI-compatible base_url | Grok/xAI, Gemini, Groq, OpenRouter, Mistral, Together, Fireworks, DeepSeek, Cohere |
| Webhook | Out-of-process platforms POST a run summary to a per-platform endpoint | Dify, n8n, Flowise, Langflow, OpenClaw |
Where to go next
- Framework integrations — LangGraph, CrewAI, OpenAI Agents, LlamaIndex, MCP, Browser Use, Google ADK, Pydantic AI.
- LLM providers — native instrumentors plus OpenAI-compatible providers.
- Agent platforms — Dify, n8n, Flowise, Langflow, Bedrock, Vertex, OpenClaw.
The in-product Connect an agent wizard generates the exact snippet or webhook URL, mints an ingest key, and gives you a Test integration button to verify the pipeline before wiring the real source.