The tool_call segment — a human-in-the-loop tool call UI

Tool call segment - rendered while waiting for a tool handler to resolve

A segment is data, not an element — it has no tag and dispatches nothing. This one is AparteToolCallSegment, and it is the object you push into a bubble; something else draws it.

Width 375 768 Full Open in a tab

Core's own renderer, drawing the example printed below inside a real viewport. A segment is data — this is what the library makes of it, and it looks the same in every framework.

Example

{
  id: 's1',
  type: 'tool_call',
  status: 'resolved',
  toolCall: { id: 'call_1', name: 'get_weather', input: { city: 'Lille' } },
  result: '11°C, overcast.',
}

Shape

Discriminated by type: 'tool_call'.

Field	Type	Required	Description
toolCall	import('./tools.js').AparteToolCall	yes
status	'pending' | 'resolved' | 'aborted' | 'awaiting-approval' | 'rejected' | 'failed'	yes	awaiting-approval — paused for a human decision; rejected — the human (or a policy) declined to run it; aborted — stopped, timed out, or nothing could run it; failed — the handler threw, and result carries the crash’s one line. More
result	string	—
structuredResult	unknown	—	The handler’s structuredContent, when it returned one — the value behind result’s prose.

Shared by every segment

From AparteSegmentBase, which is also what a segment type of your own extends.

Field	Type	Required	Description
id	string	yes	Unique segment identifier
type	string	yes	Segment type discriminator
isStreaming	boolean	—	Whether segment is currently being streamed
messageId	string	—	Id of the message this segment belongs to. Stamped on insertion.
index	number	—	Position in the owning message’s segments[], maintained across insertions and removals.
meta	Record<string, unknown> & { aparte?: AparteSegmentTiming }	—	Extras the producer of the segment knows — token counts, cost, compute device — plus aparte, the one sub-object core writes. More

Notes on individual fields

status

awaiting-approval — paused for a human decision; rejected — the human (or a policy) declined to run it; aborted — stopped, timed out, or nothing could run it; failed — the handler threw, and result carries the crash’s one line.

meta

Extras the producer of the segment knows — token counts, cost, compute device — plus aparte, the one sub-object core writes.

Everything except meta.aparte is yours: fill it with updateSegment(segmentId, { meta }), which MERGES rather than replaces. Mirrors AparteMessage.metadata.

Who emits it

An app pushes one with addSegment() on the viewport, or a parser produces it while a reply streams. Core stamps its identity and timing on insert, so an emitted segment does not have to carry them.

Who draws it

A built-in renderer draws this one, from packages/core/src/renderers/segments/tool-call.ts. It installs itself the first time a segment of this type needs it, so a chat renders it with no setup.

Replace it with your own for this config only:

import { registerSegmentRenderer } from '@aparte/core';
import type { AparteToolCallSegment } from '@aparte/core';

registerSegmentRenderer({
    type: 'tool_call',
    render: (segment: AparteToolCallSegment) => {
        const el = document.createElement('div');
        el.textContent = JSON.stringify(segment);
        return el;
    },
});

See Customization for the whole renderer seam, and bring your own loop for emitting segments yourself.