Skip to content

ActivityTrace

Stable

The agent's work shown step by step, each step tagged with who did it, then folded to one line.

Also known as activity feed, trace, steps log

Install with npm install @tattva/ui (setup), or copy the code from the kit in Downloads.

Folded, then open

  1. Done

    Read your messageAI

    You said the kettle arrived with a cracked lid.
  2. Done

    Looked up the return windowRule sheet

    Damaged items can be returned within 30 days.
  3. Done

    Worked out the refundCalculated

    ₹2,340 minus ₹0 shipping = ₹2,340
  4. Done

    Checked the amount against the orderSafety check

Folded, the trace is one line. Press it to see every step. The second trace starts open.

When to use it

Show how an answer was reached without filling the thread. The tag on each step tells the reader how much to trust it: an AI judgment can be wrong, a rule sheet lookup was checked, a calculation was done by code, and a safety check is a guardrail.

Made of: Summary button with chevron · Step list · Status icon · Step title · Actor tag · Detail

Use it for

  • An agent that reads, looks things up and calculates before it answers.
  • A thread that must stay short once the work is finished.
  • Showing which steps were the AI's judgment and which were fixed rules or code.

Not for

Do and avoid

  1. Done

    Read your messageAI

  2. Done

    Found the refund rule for damaged itemsRule sheet

  3. Done

    Worked out the reply deadlineCalculated

Do. Each tag says if the AI judged it or code worked it out.
  1. Done

    Read your messageAI

  2. Done

    Thinks the seller has 30 days to replyRule sheet

  3. Done

    Guessed you paid about ₹2,300Calculated

Avoid. AI guesses tagged Rule sheet or Calculated look more certain than they are.

Do

  • Tag every step honestly. Use AI only for judgments the model made.
  • Fold the trace when the work ends so the thread stays short.
  • Keep titles to one short line and put extra words in detail.
  • Show a failed step and say what failed.

Avoid

  • Do not tag a model guess as Rule sheet or Calculated. That would make it look more certain than it is.
  • Do not use it for steps the person must approve. Use StepTimeline or ApprovalPrompt.
  • Do not put long reasoning text in a step. Use ThinkingBlock.
  • Do not rely on the status icon colour alone. Keep the words.

On a phone

  • The summary row is at least 44px tall on touch screens and spans the full width.
  • Summaries, step titles and details wrap onto new lines instead of being cut off.
  • Opening and closing needs a tap, not hover.

Build your next AI screen with Tattva.

Free and open, with the code. Start in one line.

Get started