Traces & Audit Logs
TrustGate provides a complete "Black Box" recording of every AI interaction. Trace complex agent chains, debug errors, and export logs to your SIEM.
The Flight Recorder
A searchable immutable log of every request. Filter by AgentID, PII Detected, or Latency.
Distributed Tracing
Stitch together multi-step agent workflows (Think → Tool → Act) into a single Gantt view.
Linking Agent Steps (Tracing)
To visualize an Agent's "Chain of Thought" in the dashboard, pass the x-trustgate-trace-id header.
trace_id = str(uuid.uuid4())
# Step 1: Planning
client.chat.completions.create(
model="gpt-4",
extra_headers={
"x-trustgate-trace-id": trace_id,
"x-trustgate-workflow-step": "planning"
}
)
SIEM Export (Datadog & Splunk)
TrustGate can stream logs to your infrastructure in real-time. Configure these environment variables or set them in Settings → Observability.
Datadog
DD_API_KEYDD_SITE (e.g. datadoghq.com)Splunk
SPLUNK_HEC_URLSPLUNK_HEC_TOKENEnd-to-End Agentic Tracing
Unlike traditional LLM prompts, autonomous AI Agents often execute multi-step workflows involving tool calls, fallbacks, and internal retries. Standard single-request logging makes debugging these workflows impossible.
TrustGate introduces Agentic Tracing. We automatically group complex, multi-step agent executions into a single unified trace. This provides an end-to-end view of the agent's chain of thought, allowing you to instantly pinpoint where an agent hallucinated or got stuck in an uncontrolled retry loop.
Trace Visualization Capabilities
- Execution Timelines: View the exact millisecond duration of every tool call and sub-request within the agent's workflow.
- Retry & Hallucination Flags: TrustGate automatically flags traces where an agent excessively retries a failed tool or deviates from the allowed policy sandbox.
- Cost per Trace: Aggregate the token cost of the entire multi-step workflow, not just the individual API hits.