Data model
Open core · self-host + all Splyntra Cloud plans
Splyntra's data model is OpenTelemetry, specialized for agents. Everything you see in the dashboard is derived from OTLP spans, trace-correlated logs, and metrics — read through open semantic conventions so any OTel emitter fits without a proprietary format.
Runs vs. traces vs. spans
| Concept | Definition | Backed by |
|---|---|---|
| Run | One end-to-end agent execution — the unit you search, score, and cost. | An OTel trace, presented agent-first. |
| Trace | A tree of spans sharing a trace_id. | OpenTelemetry trace. |
| Span | A single timed operation (a step, LLM call, tool call, retrieval, or vector search). | OpenTelemetry span with attributes. |
See Traces & risk for the span kinds and the risk overlay.
Logs and metrics
The same collector ingests two more signal types, correlated with your traces:
- Logs — structured, trace-correlated log entries. Each entry auto-attaches the
active
trace_idandspan_idand is redacted with the same rules as spans, so logs line up with the trace timeline on the Logs page. Sent to/v1/logs. - Metrics — latency, throughput, error rate, and spend over time, aggregated for the Metrics screen.
Semantic conventions
Splyntra reads three families of attributes on each span. It never requires its own format — standards-compliant OTel spans ingest natively.
| Family | Purpose | Examples |
|---|---|---|
gen_ai.* | OpenTelemetry GenAI semantic conventions for model calls | model name, token usage, request/response metadata |
splyntra.* | Splyntra-specific enrichment | project, environment, framework, risk findings |
traceloop.entity.* | OpenLLMetry entity attributes, read for compatibility | workflow/entity names emitted by OpenLLMetry instrumentation |
Because the collector understands gen_ai.* and traceloop.entity.* alongside its own
splyntra.*, third-party OTel instrumentation and hyperscaler agent platforms (such as
AWS Bedrock AgentCore and Google Vertex Agent Engine) ingest at /v1/traces with no
adapter. See Ingest & OTLP.
Resource attributes
Resource attributes describe the source of the telemetry and are set once at SDK init. They power the filtering and grouping across the dashboard.
| Resource attribute | Set by | Role |
|---|---|---|
service.name | service_name / serviceName (defaults to project) | Identifies the emitting service |
splyntra.project | project | Groups runs into a project |
splyntra.environment | environment (default development) | Separates development / staging / production |
splyntra.framework | framework | Labels the agent's framework on the Agents page |
Next steps
- Traces & risk — the run/trace/span hierarchy and span kinds.
- Cost analytics — how token usage becomes cost.
- Ingest & OTLP — endpoints, headers, and the semconv on the wire.