Use case / Research agents

Show the path from question to source.

Research agents write fluent answers whether or not the evidence is there. Trace every search, source and claim so an unsupported statement is a flagged step, not a surprise.

Start tracing free Read the docs

Sources: Per claim
Citations: Checked
Gaps: Flagged

Evidence trail

Every claim needs a route back.

The run starts with the user's question, so every later step can be judged against it.
See which queries the agent planned and which it quietly dropped.
Each opened source is a span with its own timing, tokens and result.
Claims with no source are marked, instead of hiding inside a fluent paragraph.
Hold the answer, or ship it with the gap shown to the reader.

Research agents1 / 5 steps

01
Question received
“Compare vendor refund policies for 2026.”
ok
02
Search plan built
Four queries across docs and the open web.
ok
03
12 sources opened
Two paywalled, one outdated.
ok
04
3 claims unsupported
No source supports the 14-day window.
gap
05
Answer held for review
Held until the gap is visible to a person.
ok

Instrument it

Make every source a step.

Record retrievals as their own spans and the trace shows which sources each claim rests on, and which it does not.

$ npm install tracify-sdk

import { span, traceAgent } from "tracify-sdk";

export const research = traceAgent(async (question: string) => {
  const sources = await span("retrieve", () => searchWeb(question), { type: "retrieval", input: question });
  return span("synthesize", () => writeAnswer(question, sources), { type: "llm" });
}, { environment: "production" });