Skip to main content
Every conversation with Foggy is an investigation. Here’s what the experience looks like from start to finish.

​
Starting an investigation

Type a natural-language question in the chat input. You can:
  • Ask anything — “Why is the payment service returning 500 errors?”
  • @Mention resources — Type @ to reference specific Kubernetes services, deployments, or Grafana dashboards. Mentioned resources get injected as structured context for more precise investigations.
  • Choose a model — Select between Claude Sonnet 4.5, Claude Opus 4.6, and GPT-4o using the model selector in the chat input. Your selection persists within the session.
Chat input with @mention dropdown and model selector

​
What you see during an investigation

A collapsible panel labeled “Thought for Xm Ys” shows Foggy’s reasoning in real time:
  • Reasoning steps — What Foggy is thinking and what it plans to check next
  • Tool calls — The exact Grafana queries, Prometheus PromQL expressions, Loki log searches, and kubectl commands being executed
  • Plan cards — High-level investigation strategy when the question is complex
This gives you full transparency into how Foggy reached its conclusions.
After the answer, a Sources section shows all tools that were invoked during the investigation. Click any source to see the raw query and result.
The final answer streams in rich Markdown:
  • Tables summarizing findings
  • Code blocks with relevant configs or commands
  • Mermaid diagrams visualizing dependencies or timelines
  • Inline evidence linking conclusions to specific data points
Chain-of-thought expanded during an investigation

​
Follow-up suggestions

After every answer, Foggy generates 3 contextual follow-up suggestions based on the investigation results. Click any suggestion to continue the conversation and dig deeper. Examples after a latency investigation:
  • “Show the deployment diff that caused this”
  • “What other services are affected?”
  • “How do I prevent this in the future?”
Follow-up suggestions below an answer

​
Conversation history

All investigations are saved as threads in the sidebar. You can:
  • Pick up where you left off — previous messages provide context for multi-turn conversations
  • Search through past investigations
  • Share thread links with teammates

​
Message actions

Each message has actions available:
ActionWhat it does
RegenerateRe-runs the investigation with fresh data from your connectors
StopCancels a running investigation mid-stream
FeedbackThumbs up/down to help improve Foggy’s answers

​
@Mention autocomplete

Type @ in the chat input to see an autocomplete dropdown of available resources:
  • Kubernetes services and deployments
  • Grafana dashboards
  • Infrastructure components
Mentioned resources are injected as structured context, giving Foggy precise information about what you want to investigate. Model selector showing available LLMs

​
Model selection

Choose the right model for each question:
ModelBest for
Claude Sonnet 4.5Default — fast, reliable investigations
Claude Opus 4.6Complex multi-step reasoning and deep analysis
GPT-4oAlternative perspective, broad general knowledge
Selection is per-message — you can switch models mid-conversation.

​
Next steps

Slack Bot

Investigate incidents and get results directly in Slack.

Automations

Set up scheduled and alert-triggered investigations.

Knowledge Base

Add runbooks and context so Foggy gives more accurate answers.