Documentation
Docs
Architecture

Observability Layer

Consonant turns scattered agent interactions into a coherent, debuggable narrative.

The Debugging Nightmare

In traditional systems, you have stack traces. In multi-agent systems, you have non-deterministic, distributed failure modes. One agent's hallucination can cascade through five other agents, leaving you with a pile of logs that make no sense.

Zero-Code Instrumentation

With Consonant, you get full traces, token usage, and latency metrics.

Automatic Session Reconstruction

Consonant uses Global Trace IDs to stitch together every decision. When you open a session in the dashboard, we reconstruct the "Decision Tree" in real-time.

OTEL Spans

Every HTTP/gRPC request, LLM call, and tool execution is recorded as an OpenTelemetry span.

Distributed Context

Trace IDs are propagated across the network. If the "Writer Agent" fails, you can see the exact "Researcher Agent" output that caused the failure.

The Linear Narrative View

Our dashboard transforms thousands of raw spans into a readable timeline. You can "scrub" through the time-to-first-token and see exactly when each agent joined the conversation.

SESSION: d8f2-44a1-8e9c
00:01sPLANNERCreated Task: Search SEC Filings
00:04sAGENT:searcherCalling Tool: GoogleSearch(query="AAPL 10-K")
00:12sAGENT:analystEvaluating risk factors... (2,402 tokens)
00:15sERRORHallucination Detected: Invalid date format in 10-K