Skip to main content

Agent platforms

Availability

Open core · self-host + all Splyntra Cloud plans

Hosted and no-code workflow platforms run out-of-process, so they can't load an in-process SDK. Instead they POST a run summary to a per-platform webhook, and Splyntra builds a root agent span with one child span per step — redacted, validated, stored, and run through the security detectors, identical to SDK ingestion.

Webhook endpoint​

Each platform has its own receiver:

POST ${ingest}/v1/integrations/<platform>
Authorization: Bearer <ingest-key>
Content-Type: application/json
Platform<platform>Notes
DifydifyPOST the workflow_finished payload, optionally with a nodes array
n8nn8nTurnkey Splyntra community node, or a final Code + HTTP Request node
FlowiseflowiseFinal HTTP node posting the run summary
LangflowlangflowHTTP component posting the run summary
AWS Bedrock AgentCorebedrockPrefer native OTel (below); webhook for non-OTel environments
Google Vertex Agent EnginevertexPrefer native OTel (below); webhook for non-OTel environments
OpenClawopenclawSplyntra OpenClaw plugin forwards session completion

On the self-hosted collector the base is http://localhost:4318; on Cloud it is https://ingest.splyntra.com.

Test integration​

Before wiring the real source, use the Test integration button on the platform's Connect screen. It POSTs a realistic sample payload — an llm_call plus a tool_call with token counts — so you can confirm the trace, redaction, risk score, and cost land end-to-end. A representative payload looks like:

{
  "workflow": { "id": "wf_42", "name": "Support Agent" },
  "execution_id": "exec_123",
  "status": "success",
  "nodes": [
    { "name": "OpenAI", "type": "llm", "model": "gpt-4o-mini",
      "prompt_tokens": 150, "completion_tokens": 40, "elapsed_ms": 300,
      "status": "success", "output": "..." },
    { "name": "HTTP Request", "type": "tool", "elapsed_ms": 120, "status": "success" }
  ]
}

Redact input/output in your workflow before sending if you don't want raw content to leave the platform — you control exactly what is posted.

Native OTel: Bedrock & Vertex​

AWS Bedrock AgentCore and Google Vertex Agent Engine emit OpenTelemetry gen_ai.* spans. The preferred path is to point their OTLP exporter at Splyntra's trace endpoint:

POST ${ingest}/v1/traces
Authorization: Bearer <ingest-key>

Model, tokens, and cost ingest correctly with no glue, because the collector reads the gen_ai.* semantic conventions natively. Use the bedrock / vertex webhooks only for environments not wired for OTLP.

Next steps​