Trododocs
Launch AppGet Demo
Overview

Overview

Trodo is the observability layer that fixes your agents. It turns failing runs into shipped fixes: it detects what broke, explains why, and hands you the patch.

Most observability tools stop at the trace. They record everything your agent did during a run and leave you to find the needle yourself, scrolling spans for hours or burning thousands in LLM-judge bills trying to grade output at scale.

Trodo closes the loop. We turn those same runs into a self-improving agent. Trodo watches your traces for the failures that matter. Not just the crashes, but the ones that are genuinely hard to catch: hallucinations, context loss, contradictions, incompleteness, wrong format, refusals, and overreach, among many more. When a pattern crosses your threshold, Trodo tells you what broke, why, and hands you the fix, all in one place.

Observability

Everything starts with the trace. One wrap around your agent captures every run as a tree of spans: each LLM call, tool call, and nested step, with inputs, outputs, tokens, cost, latency, and errors. Works with any stack (OpenAI, Anthropic, LangChain, Vercel AI SDK, raw HTTP) through the Trodo SDK or OpenTelemetry.

  • One wrap, full trace. wrapAgent records the whole run as nested spans; auto-instrumentation captures provider calls with no extra wiring.
  • Runs and sessions. Bind runs to a user and a conversation, stitch multi-step sessions across processes and services.
  • The data layer for everything else. Issues, Evals, Capabilities, and Ask all read from the same traces.
Wrap your agent once and capture every run as a tree of spans: calls, tools, tokens, cost, and errors.

Issues

You set the thresholds; Trodo does the watching. For every failure pattern we track, you define what counts as a real problem: minimum occurrences, minimum distinct users, and a lookback window. A single weird run is noise. The same hallucination across 40 runs and 12 users in the last week is an issue worth your time.

When a pattern trips its threshold, Trodo evaluates the affected runs together, traces the failure back to your codebase, and writes up what's broken and how to fix it. Then you hit Fix this issue: a cloud coding agent (Claude Code or Cursor) implements the fix, runs your tests, and pushes a branch. Review it, then open a PR straight from Trodo into prod or any branch you choose.

  • Patterns across five families: tool failures, output quality (hallucination, contradiction, off-topic, incompleteness, wrong format), tool misuse, conversation breakdown (context loss, agent loops, rephrase loops), and UX rage.
  • Threshold-gated, not alert spam. Issues only surface once they clear the volume and user bar you set.
  • Grouped and root-caused. Each issue bundles its member runs, the failing spans, and a plain-language explanation of why it happened.
  • Fix this issue, in one click. A coding agent writes the fix, runs your tests, and pushes a branch; you open the PR from Trodo.
Set thresholds, watch issues surface with their root cause, then fix them in one click.

Ask

Most observability tools make you the query engine: filter, scroll, read, repeat. Trodo makes everything queryable in plain language. Ask for the last 25 runs from a single user, reconstruct a journey across your product, or pull a report, and Trodo plans the investigation, runs the right queries, and answers.

  • Natural-language investigation. Describe what you want to know; Trodo finds it across runs, spans, and events.
  • Scoped to anything: a user, a session, a cohort, a board, or a single run.
  • Built for analysis, not just search. Summaries, journeys, and reports, not raw rows you have to interpret.
Query traces, runs, and journeys in plain language. No filters, no SQL.

Evals

Measure agent quality your way, with four kinds of evaluators that run continuously over your runs and spans.

  • Python. Write code that takes a run or span and returns pass/fail with metadata.
  • LLM-as-judge. Bring your own model, define a rubric, grade output at scale.
  • Human. Route runs into a grading queue for human verdicts.
  • Composite. Combine multiple evaluators with AND/OR logic into one score.

Test any evaluator one-shot against a single run, then backfill it across your history to see how you've been doing all along.

Python, LLM-as-judge, human, and composite evaluators, running continuously over every run.

Trodo Agents

Build agents on top of your own Trodo data. The visual workflow editor lets you wire your traces, events, and metrics into agents that connect to your apps, your tools, and other agents, triggered by webhooks or on a schedule, with credentials and versioning managed for you.

  • Your data as fuel. Agents read from the traces and metrics Trodo already holds.
  • Connected. Wire in external tools and MCP servers as dynamic capabilities.
  • Versioned. Draft, publish, and roll back workflow versions safely.
Build agents from your Trodo data and connect them to your apps and other agents.

Playground

Test and compare prompts and models before they ship. Run candidates side by side, measure quality at scale with your evals, and pick the best, without guessing.

  • Prompt and model comparison. Try variations against the same inputs.
  • Quality at scale. Score candidates with your existing evaluators, not vibes.
  • Ship with confidence. Promote the winner once the numbers back it.

Capabilities

Trodo clusters your agent runs by what users are actually trying to do, then scores how well each cluster is served. You see, at a glance, the jobs your agent handles well and the ones where it falls short.

  • Automatic clustering. Runs group by intent, no manual tagging.
  • Satisfaction per cluster. Each capability carries a score from red to green.
  • Capability gaps. See where users ask for something your agent can't yet do.
Cluster runs by intent, score satisfaction per cluster, and find the gaps in what your agent can do.

Product Analytics

The human side of your product. Track what users actually do across your surface, then slice it with funnels, flows, retention, and cohorts. Agent runs and product events share the same user identity, so you can move between them in one place.

  • Events and users. Auto-captured behavior plus your custom events, tied to user profiles.
  • Funnels, flows, retention. Understand activation, drop-off, and what brings users back.
  • Shared identity. Segment agent behavior by product behavior and the reverse.
Track product behavior with events, funnels, flows, and retention, all tied to the same user identity as your runs.

Reports

Build the boards and dashboards your team lives in. Compose insights, funnels, flows, and retention into saved reports, pin the ones that matter, and share them.

  • Boards and insights. Assemble multiple visualizations into one view.
  • Pin and share. Keep key dashboards one click away, share public links.
  • Versioned. Reports carry history so you can restore a prior state.
Compose insights into boards, pin what matters, and share dashboards across your team.

Get started

Agent Observability

Wrap your agent and see your first trace in under 2 minutes.

Product Analytics

Install the SDK and track your first event in under 5 minutes.

Concepts

Understand runs, spans, identity, and how it all fits together.

Start Tracing

Instrument your agent and send your first runs to Trodo.

On this page

ObservabilityIssuesAskEvalsTrodo AgentsPlaygroundCapabilitiesProduct AnalyticsReportsGet started