Skip to main content

Documentation Index

Fetch the complete documentation index at: https://whyops.com/docs/llms.txt

Use this file to discover all available pages before exploring further.

The WhyOps frontend is a Next.js 15 application designed to visualize and analyze the complex, non-deterministic decision paths of autonomous agents.

​
Core Visualizations

​
1. Decision Graphs

Instead of linear logs, WhyOps uses a directed acyclic graph (DAG) to represent the flow of an agent’s reasoning, tool execution, and memory retrieval.
  • Technology: Built using reactflow for the interactive node-link diagrams and dagre for automated layout and routing.
  • Node Types: Represents specific events in a trace, such as a user prompt, an LLM call, a tool execution, or a failure.
  • Edges: Represents causality and data flow between nodes (e.g., the output of a tool call feeding into the next LLM prompt).

​
2. Trace Inspector

When a user clicks on a node in the decision graph, a detailed inspector pane opens, revealing the precise payload of that event.
  • Payloads: View raw JSON of prompts, completions, and tool arguments.
  • Metadata: See latency, token usage, and the specific model version used.
  • Diffs: Compare changes in agent state or memory across sequential steps using react-diff-viewer-continued.

​
3. Agent Knowledge Profiles

A high-level dashboard that aggregates the results of automated evaluations (whyops-analyse) across all runs of a specific agent.
  • Metrics: Track average scores on dimensions like intent_precision and reliability_recovery over time.
  • Visuals: Uses recharts to render trend lines and distributions of agent performance.
  • Insights: Highlight recurring failure modes (e.g., “This agent frequently enters an infinite loop when the search_docs tool returns no results”).

​
State Replay (Coming Soon)

A core tenet of WhyOps is the ability to recreate the exact environment in which an agent failed. The frontend is building tools to extract the state of an agent at a specific node in the decision graph and replay it locally, allowing developers to debug failures without re-running expensive or side-effect-heavy steps.