Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.trodo.ai/docs/llms.txt

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

Trodo’s in-app chat is a natural-language interface to the same data the MCP server exposes — but with answers rendered as interactive charts, run inspectors, profile cards, and report builders inline, not as JSON.

​
What you can ask

A few examples (the chat handles ~70 tools across analytics, agent runs, evals, issues, identity, UX/health, reports, and catalog discovery):
  • Analytics — “yesterday’s signup funnel”, “DAU last 7 days”, “retention curve for new users last month”, “compare DAU before and after we shipped X”, “top 10 most active users this month”.
  • Agent runs — “show recent agent runs”, “details for run abc-123”, “p95 latency for the support agent”, “what users tried to checkout but got an error”, “summarize cluster 7”.
  • Evals — “list our evaluators and their pass rates”, “show evaluator config for helpfulness”, “eval results for run abc-123”, “users who failed eval helpfulness this week”, “any pending human evals to grade”.
  • Eval lifecycle (write) — “create a new evaluator for compliance signals”, “disable the helpfulness evaluator”, “test evaluator helpfulness on run abc-123”, “backfill compliance evaluator over the last 14 days”.
  • Issues — “list open issues sorted by severity”, “details for issue 41”, “linked runs for issue 41”, “top failing tools this week”, “rage hotspots on the pricing page”, “acknowledge issue 41”.
  • Identity — “show me [email protected] profile”, “find users named alice”, “lookup wallet 0xdeadbeef”, “users who fired purchase_completed last 14 days”, “who is impacted by issue 41”.
  • UX & health — “rage clicks on pricing page”, “JS errors affecting the most users”, “core web vitals for checkout”, “network errors by status code”, “form abandonment on signup”.
  • Reports — “create a report on this week’s checkout funnel”, “save these results as a report”.
  • Cross-domain — “find [email protected] and show her recent agent runs”, “what runs failed eval helpfulness yesterday”, “users impacted by the top failing tool this week”.
The chat resolves natural language to IDs server-side — pass emails, names, free-text run / cluster / issue references; the right UUIDs are looked up automatically.

​
How answers come back

  • Visual by default. Any multi-row numeric result (funnel steps, retention cohorts, time series, breakdowns, ranked lists) renders as an interactive chart inline. Single-number answers stay as text.
  • Right-sized. Lookup / metric / ranking questions return data only — no unsolicited “Insights” / “Suggestions” / “Follow-ups” sections. Ask “why X is happening” to get analysis. Ask “how do I fix X” or “what should I do” to get recommendations. Ask for a “report” to get an exportable dashboard preview.
  • Interactive widgets. Tool results that have a natural rendering (agent runs, user profiles, issues, eval forms, reports) render as cards you can drill into — no JSON dumps.
  • Discovery-first. Before passing event / property / agent names to analytical tools, the chat checks the team’s catalog so it never guesses names that don’t exist.
  • Grounded. Numbers only appear in answers when they came from a tool result. The synthesizer cannot invent figures or statistics.

​
Privacy

All queries are scoped to the team bound to your account. The chat reads the same data the MCP server does — anything in MCP is also in chat (and vice versa).