Incident Management
Platform
AI
Dedicated compute
Dedicated compute keeps concurrent queries stable and complex analysis fast.
Learn moreDedicated compute
Dedicated compute keeps concurrent queries stable and complex analysis fast.
Learn moreAI Agent Observability goes beyond checking whether a model endpoint returned successfully. It connects Prompt and Completion data with model calls, tool execution, retrieval, guardrails, latency, Token usage, and the business workflow around each Session and Trace.
Create a common operational view across multi-turn conversations, model calls, tool execution, Token cost, and risk events for supported Agent runtimes and custom enterprise applications.
OpenClaw
Hermes
Claude Code
Codex
Other AI AgentsAgent execution is a black box:A final response does not explain the intermediate model, retrieval, and tool steps that produced it.
Token cost and latency are difficult to control:Long context, retries, and repeated tool output can make one task unexpectedly expensive or slow.
Risk events are hard to locate:A dangerous command or sensitive-data action must be tied back to the exact Session, Trace, and Span.
AI workflows are disconnected from production systems:Without shared service and business context, teams cannot measure downstream impact or ownership.
Session and Trace analysis:Reconstruct each conversation, turn, and execution path with stable identifiers.
Model and tool-call analysis:Compare requests, arguments, outputs, status, duration, and call sequence.
Token and performance monitoring:Measure input, output, cached Token use, latency, retries, and cost drivers.
Risk-event correlation:Link guardrail and behavioural events to the exact action and surrounding context.
Search sessions by Agent, instance, model, Token range, time, and risk level. Open the affected session to review every turn and identify the Trace where behaviour, latency, or cost diverged.
Inspect model calls, assistant steps, tool execution, status, duration, and risk events in sequence. This shows whether the problem came from model reasoning, generated arguments, tool permissions, or the target system.
Use Span timing and waterfall-style execution views to compare model latency, retrieval, tool time, retries, and orchestration overhead. Find the step that dominates latency or repeatedly consumes Tokens without progress.
Attach Agent type, instance, model provider, request model, Session ID, Trace ID, and business attributes during instrumentation. Consistent identity makes performance, cost, quality, and risk analysis reusable across teams.
LLM Observability focuses on model requests, responses, latency, Token usage, and quality. AI Agent Observability extends that view to multi-turn Sessions, orchestration, retrieval, tool calls, state changes, and risk events across the full Agent workflow.
Depending on instrumentation, Guance can analyse Session and Trace identity, model provider and model, input and output Tokens, latency, Span hierarchy, tool calls, arguments, status, retries, risk events, and custom business attributes.
Create an Agent or LLM monitoring application, instrument the runtime with supported OpenTelemetry and GenAI semantic attributes or the documented integration, and preserve stable Session, Trace, Agent, model, and business identifiers.
AI engineers, application developers, platform teams, SREs, security teams, and product owners use it to improve Agent reliability, performance, cost control, quality, and operational accountability.
Build a useful operational view of model requests, latency, errors, and cost.
Create an application, instrument an Agent, and analyse Session and Trace evidence.
Bring logs, metrics, traces, events, RUM, and infrastructure queries into reviewable Agent tasks.
Connect supported AI clients to authorised Guance context through Model Context Protocol.
