Skip to content

armgit / avibe

Public Imported from GitHub
0

Make your computer a home for your AI partner. Avibe is the local-first Agent OS: Vibey, its built-in AI partner, works out of the box, and Claude Code, Codex & OpenCode run unchanged beside it — reach them from your browser, phone, or any chat app.

feat(models): check image generation like the other tools, total a set's cost without counting unknown as zero, and show a replay that did not verify (#2624)

Call Whitee41569c8381a
diff --git a/tests/scenarios/model_hub/catalog.yaml b/tests/scenarios/model_hub/catalog.yaml
index 926a2a4ed..cf8f2cece 100644
--- a/tests/scenarios/model_hub/catalog.yaml
+++ b/tests/scenarios/model_hub/catalog.yaml
@@ -445,6 +445,7 @@ scenarios:
- 'ui/src/components/settings/models/SourceDetailPanel.test.tsx::MH-HOSTED-UI-003: says a re-check''s cost is unknown when no estimate is known'
- 'ui/src/components/settings/models/hostedTools.test.ts::MH-HOSTED-UI-003: totals a check only from what the server estimates, and never counts unknown as zero'
- 'ui/src/components/settings/models/format.test.ts::MH-HOSTED-UI-003: keeps the first significant digits of a per-call price, and never shows a nonzero price as zero'
+ - 'ui/src/components/settings/models/ToolsTab.test.tsx::MH-HOSTED-UI-003: lets the locale compose a set''s cost with its unpriced checks'
- id: MH-HOSTED-UI-004
name: The Tools tab says how many checks an engine upgrade made stale, and nothing when none did
status: covered
@@ -499,7 +500,7 @@ scenarios:
- 'ui/src/components/settings/models/SourceDetailPanel.test.tsx::MH-HOSTED-UI-013: checks a tool again from its chip, with the cost on the action, and answers there when the check is not evidence'
- 'ui/src/components/settings/models/useHostedTools.test.tsx::MH-HOSTED-UI-013: reads the record before calling a probe unanswered when its answer was lost'
- 'ui/src/components/settings/models/useHostedTools.test.tsx::MH-HOSTED-UI-013: reads a verdict the status does not carry as not verified, with why'
- - 'ui/src/components/settings/models/hostedTools.test.ts::MH-HOSTED-UI-013: offers to check every pair with no current answer, and never an image'
+ - 'ui/src/components/settings/models/hostedTools.test.ts::MH-HOSTED-UI-013: offers to check every pair with no current answer, images included'
- id: MH-HOSTED-UI-016
name: A hosted-tool check is a real call on exactly one Source, so the Models page reads its Sources again after one, and a Source the probe verified shows as verified
status: covered
@@ -509,7 +510,7 @@ scenarios:
- 'ui/src/components/settings/models/SettingsModelsPage.render.test.tsx::MH-HOSTED-UI-016: reads the hosted-tools report again after a committed Source change, since its models decide which pairs could run a tool'
- 'ui/src/components/settings/models/SettingsModelsPage.render.test.tsx::MH-HOSTED-UI-016: reads the Sources again after a check, since a probe can verify its Source'
- id: MH-HOSTED-UI-018
- name: The Tools tab never enables a tool that cannot run. It stays off until the server binds an executor, says whether a check or an executor pick is missing or checks are not available yet, and an enabled tool with nothing verified says its calls fail
+ name: The Tools tab never enables a tool that cannot run. It stays off until the server binds an executor, says whether a check or an executor pick is missing, and an enabled tool with nothing verified says its calls fail
status: covered
kind: contract
layer: unit
@@ -517,7 +518,7 @@ scenarios:
- 'ui/src/components/settings/models/ToolsTab.test.tsx::MH-HOSTED-UI-018: cannot turn a tool on until a pair is verified to run it, but can always turn it off'
- 'ui/src/components/settings/models/ToolsTab.test.tsx::MH-HOSTED-UI-018: keeps a tool off while the server binds no executor for it, and says to choose one when a verified pair exists'
- 'ui/src/components/settings/models/ToolsTab.test.tsx::MH-HOSTED-UI-018: claims routing only while the tool is on, and shows an off tool''s executor as a saved preference'
- - 'ui/src/components/settings/models/ToolsTab.test.tsx::MH-HOSTED-UI-018: says plainly when a tool cannot be checked yet, instead of asking for a check it cannot run'
+ - 'ui/src/components/settings/models/ToolsTab.test.tsx::MH-HOSTED-UI-018: keeps an image tool off until a check verifies one of its pairs, and offers that check'
- 'ui/src/components/settings/models/ToolsTab.test.tsx::MH-HOSTED-UI-018: says an enabled tool with nothing verified fails its calls, instead of naming a proxy'
- id: MH-HOSTED-UI-017
name: A Source's capability chip is usable without a mouse. Its spoken name is the locale's composition of the label, explanation and facts, it announces the detail it opens, and Enter moves focus to the detail's paid re-check
@@ -527,6 +528,12 @@ scenarios:
test:
- 'ui/src/components/settings/models/SourceDetailPanel.test.tsx::MH-HOSTED-UI-017: names the record time, how it was learned, the reason, and until when it holds'
- 'ui/src/components/settings/models/SourceDetailPanel.test.tsx::MH-HOSTED-UI-017: moves keyboard focus into the detail on Enter, so its paid re-check can be reached'
+ - id: MH-HOSTED-UI-019
+ name: A check's replay step reads as its own result. A replay that ran and did not verify asks for another try (or says the Source does not support it), and no replay says nothing; neither reads as the call failing
+ status: covered
+ kind: observation
+ layer: unit
+ test: ui/src/components/settings/models/ToolsTab.test.tsx::MH-HOSTED-UI-019
- id: MH-VIBEY-007
name: Vibey starts once with the user's Sources and built-in Agent models, seeded when its first Source exists
status: covered
diff --git a/ui/src/components/settings/models/HostedCapabilityChips.tsx b/ui/src/components/settings/models/HostedCapabilityChips.tsx
index b3155738f..fdb77e920 100644
--- a/ui/src/components/settings/models/HostedCapabilityChips.tsx
+++ b/ui/src/components/settings/models/HostedCapabilityChips.tsx
@@ -14,7 +14,7 @@ import { Button } from '@/components/ui/button';
import { Popover, PopoverAnchor, PopoverContent } from '@/components/ui/popover';
import { cn } from '@/lib/utils';
import { formatDayTime, formatMonthDay, formatSentences } from './format';
-import { probeableTool, probeKey, type ModelCapability, type ProbeTarget } from './hostedTools';
+import { probeKey, type ModelCapability, type ProbeTarget } from './hostedTools';
import type { HostedToolName, HostedToolProbeResult, RefusedEffort } from './types';
import { ProbeAnswerNote } from './HostedProbeDialog';
import { useEstimateLabel } from './useEstimateLabel';
@@ -221,7 +221,7 @@ const CapabilityChip: React.FC<{
tool,
toolInline: t(`settings.models.tools.tool.${capability.tool}.inline`),
}) as string;
- const action = onProbe && probeableTool(capability.tool)
+ const action = onProbe
? (
<span className="flex flex-col items-end gap-1">
{last !== undefined && <ProbeAnswerNote result={last} />}
diff --git a/ui/src/components/settings/models/HostedProbeDialog.tsx b/ui/src/components/settings/models/HostedProbeDialog.tsx
index 87c9f163f..f9aae17d2 100644
--- a/ui/src/components/settings/models/HostedProbeDialog.tsx
+++ b/ui/src/components/settings/models/HostedProbeDialog.tsx
@@ -11,10 +11,10 @@ import { AlertCircle, AtSign, Check, Globe, Image as ImageIcon, Loader2, RotateC
import { Button } from '@/components/ui/button';
import { Dialog, DialogContent, DialogDescription, DialogTitle } from '@/components/ui/dialog';
import { cn } from '@/lib/utils';
-import { probeKey, probeTotal, probeUnanswered, type ProbeTarget } from './hostedTools';
+import { probeCostSummary, probeKey, probeUnanswered, type ProbeTarget } from './hostedTools';
import { serverText } from './serverCopy';
import type { HostedToolName, HostedToolProbeResult, Source } from './types';
-import { useEstimateLabel } from './useEstimateLabel';
+import { useCostSummaryLabel, useEstimateLabel } from './useEstimateLabel';
const TOOL_ICON: Record<HostedToolName, React.ComponentType<{ className?: string }>> = {
web_search: Globe,
@@ -70,7 +70,7 @@ const OutcomeChip: React.FC<{ outcome: Outcome }> = ({ outcome }) => {
*/
export const ProbeAnswerNote: React.FC<{ result: HostedToolProbeResult | null }> = ({ result }) => {
const { t } = useTranslation();
- if (!probeUnanswered(result)) return null;
+ if (!probeUnanswered(result)) return result ? <ReplayNote replay={result.replay} /> : null;
const failure = serverText(t, result?.call.error);
return (
<span className="model-hub-tool-cost-note flex items-center gap-1 model-hub-ink-gold" role="status">
@@ -84,6 +84,22 @@ export const ProbeAnswerNote: React.FC<{ result: HostedToolProbeResult | null }>
);
};
+/**
+ * The replay step a verified call is followed by, when it ran and did not verify.
+ * Null (no replay ran) and a verified replay say nothing; a replay that got no
+ * answer asks for another try, never reads as a failure.
+ */
+const ReplayNote: React.FC<{ replay: HostedToolProbeResult['replay'] }> = ({ replay }) => {
+ const { t } = useTranslation();
+ if (!replay || replay.outcome === 'verified') return null;
+ return (
+ <span className="model-hub-tool-cost-note flex items-center gap-1 model-hub-ink-gold" role="status">
+ <AlertCircle className="size-3 shrink-0" aria-hidden />
+ {t(replay.outcome === 'unsupported' ? 'settings.models.tools.probe.replay.unsupported' : 'settings.models.tools.probe.replay.unverified')}
+ </span>
+ );
+};
+
/** Mounted for one check: the set it shows is the one it was opened with. */
export const HostedProbeDialog: React.FC<{
targets: readonly ProbeTarget[];
@@ -93,6 +109,7 @@ export const HostedProbeDialog: React.FC<{
}> = ({ targets, sources, onRun, onClose }) => {
const { t } = useTranslation();
const estimateLabel = useEstimateLabel();
+ const costSummaryLabel = useCostSummaryLabel();
// The set is fixed when the dialog opens: the read refreshes under it as probes land.
const [planned, setPlanned] = React.useState<readonly ProbeTarget[]>(targets);
const [phase, setPhase] = React.useState<'confirm' | 'running' | 'result'>('confirm');
@@ -107,13 +124,12 @@ export const HostedProbeDialog: React.FC<{
setPhase('result');
};
- const total = probeTotal(planned);
- const totalLabel = estimateLabel(total);
+ const totalLabel = costSummaryLabel(probeCostSummary(planned));
const outcomes = planned.map((target) => outcomeOf(results.get(probeKey(target)) ?? null));
const unverified = planned.filter((target) => results.has(probeKey(target)) && probeUnanswered(results.get(probeKey(target)) ?? null));
const count = (wanted: (outcome: Outcome) => boolean) => outcomes.filter(wanted).length;
// A retry is as paid as the first run, so it says its cost the same way.
- const retryLabel = estimateLabel(probeTotal(unverified));
+ const retryLabel = costSummaryLabel(probeCostSummary(unverified));
const done = results.size;
return (
@@ -144,7 +160,7 @@ export const HostedProbeDialog: React.FC<{
<span className="font-semibold text-foreground">{t('settings.models.tools.dialog.total')}</span>
<span className="text-muted">{t('settings.models.tools.dialog.totalNote')}</span>
</span>
- <span className="font-bold text-foreground">{totalLabel ?? t('settings.models.tools.probe.unknown')}</span>
+ <span className="font-bold text-foreground">{totalLabel}</span>
</div>
</>
)}
@@ -181,6 +197,7 @@ export const HostedProbeDialog: React.FC<{
: <Loader2 className="size-3.5 animate-spin text-muted" aria-label={t('settings.models.tools.probe.running') as string} />}
/>
{results.has(key) && reason && <p className="model-hub-probe-reason">{reason}</p>}
+ {results.has(key) && !reason && result && <ReplayNote replay={result.replay} />}
</React.Fragment>
);
})}
@@ -194,7 +211,7 @@ export const HostedProbeDialog: React.FC<{
<>
<Button variant="outline" size="sm" onClick={onClose}>{t('settings.models.tools.dialog.cancel')}</Button>
<Button size="sm" disabled={planned.length === 0} onClick={() => void run(planned)}>
- {totalLabel ? t('settings.models.tools.dialog.startCost', { cost: totalLabel }) : t('settings.models.tools.dialog.start')}
+ {t('settings.models.tools.dialog.startCost', { cost: totalLabel })}
</Button>
</>
)}
@@ -207,7 +224,7 @@ export const HostedProbeDialog: React.FC<{
void run(unverified);
}}>
<RotateCw className="size-3.5" aria-hidden />
- {t('settings.models.tools.result.retryCost', { count: unverified.length, cost: retryLabel ?? t('settings.models.tools.probe.unknown') })}
+ {t('settings.models.tools.result.retryCost', { count: unverified.length, cost: retryLabel })}
</Button>
)}
<Button size="sm" onClick={onClose}>{t('settings.models.tools.result.done')}</Button>
diff --git a/ui/src/components/settings/models/ToolsTab.test.tsx b/ui/src/components/settings/models/ToolsTab.test.tsx
index a9b5bcc2f..6297ad271 100644
--- a/ui/src/components/settings/models/ToolsTab.test.tsx
+++ b/ui/src/components/settings/models/ToolsTab.test.tsx
@@ -365,36 +365,61 @@ describe('ToolsTab checks', () => {
within(card('Web search').querySelector('[data-candidate="src_grok0001:grok-4.6"]') as HTMLElement).getByRole('button', { name: 'Check again · about $0.01' });
within(card('Web search').querySelector('[data-candidate="src_chatgpt1:gpt-6.1-sol"]') as HTMLElement)
.getByRole('button', { name: 'Check · included in subscription' });
- // Images cannot be checked yet, so their rows offer nothing to run.
- expect(within(card('Image generation')).queryByRole('button', { name: /Check/ })).toBeNull();
+ // An image check has no list price: it says so, and never shows $0.
+ within(card('Image generation').querySelector('[data-candidate="src_relay001:gpt-6.1-sol"]') as HTMLElement)
+ .getByRole('button', { name: 'Check · cost unknown' });
});
it('MH-HOSTED-UI-013: runs the header check only after the dialog shows what it sends and costs', async () => {
- const onProbe = vi.fn<RunProbe>(async (target) => target.candidate.source_id === RELAY.id
+ const onProbe = vi.fn<RunProbe>(async (target) => target.tool === 'web_search' && target.candidate.source_id === RELAY.id
? result('verified', target.candidate.source_id, target.candidate.model_id)
: { ...result('transient', target.candidate.source_id, target.candidate.model_id), call: { outcome: 'transient', error: 'models.source.cooldown.rate_limited' } });
renderTab(searchable(), { onProbe });
- // Two unchecked pairs (the verified one and the image are not offered); the subscription adds nothing.
- screen.getByText('2 pairs to check · about $0.02 in total');
+ // Three unchecked pairs (the verified one is not offered): the priced one is summed,
+ // the included and the unpriced image are counted, and nothing is added as $0.
+ screen.getByText('3 pairs to check · about $0.02 + 1 included + 1 without a price in total');
await userEvent.click(screen.getByRole('button', { name: 'Check available tools' }));
const dialog = await screen.findByRole('dialog');
- within(dialog).getByText('This sends 2 real requests to the Sources below to confirm they can run these tools natively.');
+ within(dialog).getByText('This sends 3 real requests to the Sources below to confirm they can run these tools natively.');
expect(onProbe).not.toHaveBeenCalled();
- await userEvent.click(within(dialog).getByRole('button', { name: 'Start check · about $0.02' }));
+ await userEvent.click(within(dialog).getByRole('button', { name: 'Start check · about $0.02 + 1 included + 1 without a price' }));
await within(dialog).findByText('Check finished');
- expect(onProbe).toHaveBeenCalledTimes(2);
+ expect(onProbe).toHaveBeenCalledTimes(3);
within(dialog).getByText('1 runs natively');
- within(dialog).getByText('1 not verified');
+ within(dialog).getByText('2 not verified');
// The reason sits where the locale puts it in the sentence.
- within(dialog).getByText('Couldn\'t verify this time (Rate limited). Try again; this is not recorded as unsupported.');
+ expect(within(dialog).getAllByText('Couldn\'t verify this time (Rate limited). Try again; this is not recorded as unsupported.')).toHaveLength(2);
- // Retrying runs only the pair that got no answer.
- await userEvent.click(within(dialog).getByRole('button', { name: 'Retry 1 · included in subscription' }));
+ // Retrying runs only the pairs that got no answer, with their cost said the same way.
+ await userEvent.click(within(dialog).getByRole('button', { name: 'Retry 2 · included in subscription + 1 without a price' }));
await within(dialog).findByText('Check finished');
- expect(onProbe).toHaveBeenCalledTimes(3);
- expect(onProbe.mock.calls[2][0]).toMatchObject({ candidate: { source_id: SUBSCRIPTION.id } });
+ expect(onProbe).toHaveBeenCalledTimes(5);
+ expect(onProbe.mock.calls.slice(3).map(([target]) => target.candidate.source_id).sort()).toEqual([RELAY.id, SUBSCRIPTION.id].sort());
+ });
+
+ it('MH-HOSTED-UI-003: lets the locale compose a set\'s cost with its unpriced checks', () => {
+ renderTab(searchable(), { lng: 'zh' });
+ screen.getByText('3 项可检测 · 共约 $0.02,另 1 项订阅内含,另 1 项暂无单价');
+ });
+
+ it('MH-HOSTED-UI-019: shows a replay that did not verify as its own result, and nothing when no replay ran', async () => {
+ const replayed = (replay: HostedToolProbeResult['replay']): HostedToolProbeResult => ({ ...result('verified', RELAY.id, 'gpt-6-astra'), replay });
+ const onProbe = vi.fn<RunProbe>()
+ .mockResolvedValueOnce(replayed({ outcome: 'transient', error: null }))
+ .mockResolvedValueOnce(replayed(null));
+ renderTab(searchable(), { onProbe });
+ const row = card('Web search').querySelector('[data-candidate="src_relay001:gpt-6-astra"]') as HTMLElement;
+
+ // The call verified; the replay after it ran out of time. That asks for another try, never reads as a failure.
+ await userEvent.click(within(row).getByRole('button', { name: 'Check · about $0.02' }));
+ within(row).getByText('Replay not verified · try again');
+ expect(row.textContent).not.toMatch(/Not verified|Couldn't verify/);
+
+ // No replay ran: nothing more to say.
+ await userEvent.click(within(row).getByRole('button', { name: /^Check/ }));
+ expect(within(row).queryByText(/Replay/)).toBeNull();
});
it('MH-HOSTED-UI-013: says why a failed check failed, and never shows the server\'s key as text', async () => {
@@ -413,12 +438,12 @@ describe('ToolsTab checks', () => {
expect(row.textContent).not.toContain('models.source');
});
- it('MH-HOSTED-UI-018: says plainly when a tool cannot be checked yet, instead of asking for a check it cannot run', () => {
+ it('MH-HOSTED-UI-018: keeps an image tool off until a check verifies one of its pairs, and offers that check', () => {
renderTab(status([tool('image_generation', { candidates: [{ source_id: RELAY.id, model_id: 'gpt-6.1-sol', call: unknown, replay: null }] })]));
const image = card('Image generation');
- within(image).getByText('Off · checks for this tool come in a later update');
+ within(image).getByText('Off · check a Source first');
expect((within(image).getByRole('switch') as HTMLButtonElement).disabled).toBe(true);
- expect(within(image).queryByRole('button', { name: /Check/ })).toBeNull();
+ within(image).getByRole('button', { name: 'Check · cost unknown' });
});
it('MH-HOSTED-UI-003: says a check\'s cost is unknown before it runs, when no estimate is known', () => {
diff --git a/ui/src/components/settings/models/ToolsTab.tsx b/ui/src/components/settings/models/ToolsTab.tsx
index f7098b090..af7e9496e 100644
--- a/ui/src/components/settings/models/ToolsTab.tsx
+++ b/ui/src/components/settings/models/ToolsTab.tsx
@@ -19,9 +19,8 @@ import { HostedProbeDialog, ProbeAnswerNote } from './HostedProbeDialog';
import {
evidenceChip,
pendingChecks,
- probeableTool,
probeKey,
- probeTotal,
+ probeCostSummary,
probeUnanswered,
sameExecutor,
verifiedCandidates,
@@ -40,7 +39,7 @@ import type {
HostedToolStatus,
Source,
} from './types';
-import { useEstimateLabel } from './useEstimateLabel';
+import { useCostSummaryLabel, useEstimateLabel } from './useEstimateLabel';
type RunProbe = (target: ProbeTarget) => Promise<HostedToolProbeResult | null>;
@@ -232,7 +231,7 @@ const ToolCard: React.FC<{
? 'settings.models.tools.state.off'
: pickable
? 'settings.models.tools.state.needsExecutor'
- : probeableTool(tool.name) ? 'settings.models.tools.state.needsCheck' : 'settings.models.tools.state.checkUnavailable')}</span>
+ : 'settings.models.tools.state.needsCheck')}</span>
<ToggleSwitch
enabled={tool.enabled}
disabled={pending || locked}
@@ -260,9 +259,7 @@ const ToolCard: React.FC<{
candidate={candidate}
source={sources.get(candidate.source_id)}
executor={tool.enabled && sameExecutor(candidate, bound)}
- probe={probeableTool(tool.name)
- ? { running: probing.has(key), last: lastProbes.get(key), run: onProbe, disabled: probing.size > 0 && !probing.has(key) }
- : null}
+ probe={{ running: probing.has(key), last: lastProbes.get(key), run: onProbe, disabled: probing.size > 0 && !probing.has(key) }}
/>
);
})}
@@ -331,7 +328,7 @@ export const ToolsTab: React.FC<{
onRetry: () => void;
}> = ({ read, sources, pending, probing, onChange, onProbe, onRetry }) => {
const { t } = useTranslation();
- const estimateLabel = useEstimateLabel();
+ const costSummaryLabel = useCostSummaryLabel();
// A row's last answer that did not become evidence (inconclusive, transient, a
// failed request) shows where it was asked, until that row is checked again.
const [lastProbes, setLastProbes] = React.useState<ReadonlyMap<string, HostedToolProbeResult | null>>(() => new Map());
@@ -352,13 +349,13 @@ export const ToolsTab: React.FC<{
const reverify = status?.reverify;
const toCheck = status ? pendingChecks(status) : [];
// 「全部已检测」 holds only where there is something a check could run on.
- const checkable = status?.tools.some((tool) => probeableTool(tool.name) && tool.candidates.length > 0) ?? false;
- const toCheckTotal = probeTotal(toCheck);
- const toCheckCost = toCheckTotal === null
+ const checkable = status?.tools.some((tool) => tool.candidates.length > 0) ?? false;
+ const toCheckSummary = probeCostSummary(toCheck);
+ const toCheckCost = toCheckSummary.priced === 0 && toCheckSummary.included === 0
? t('settings.models.tools.header.totalUnknown')
- : toCheckTotal.basis === 'included'
+ : toCheckSummary.priced === 0 && toCheckSummary.unpriced === 0
? t('settings.models.tools.header.totalIncluded')
- : t('settings.models.tools.header.total', { cost: estimateLabel(toCheckTotal) });
+ : t('settings.models.tools.header.total', { cost: costSummaryLabel(toCheckSummary) });
return (
<section className="model-hub-tools flex flex-col" aria-label={t('settings.models.tools.title') as string}>
<div className="flex flex-col gap-2 sm:flex-row sm:items-end sm:justify-between">
diff --git a/ui/src/components/settings/models/hostedTools.test.ts b/ui/src/components/settings/models/hostedTools.test.ts
index b240da2ee..4e9cbc585 100644
--- a/ui/src/components/settings/models/hostedTools.test.ts
+++ b/ui/src/components/settings/models/hostedTools.test.ts
@@ -1,6 +1,6 @@
import { describe, expect, it } from 'vitest';
-import { evidenceChip, modelCapabilities, pendingChecks, probeTotal, type ProbeTarget } from './hostedTools';
+import { evidenceChip, modelCapabilities, pendingChecks, probeCostSummary, type ProbeTarget } from './hostedTools';
import type { HostedEvidence, HostedToolCandidate } from './types';
const evidence = (overrides: Partial<HostedEvidence>): HostedEvidence => ({
@@ -51,15 +51,14 @@ describe('hosted-tool evidence projection', () => {
candidate: { ...candidate('src_relay001', 'gpt-6-astra', evidence({})), probe_estimate: estimate },
});
- expect(probeTotal([at({ basis: 'list', usd: 0.02 }), at({ basis: 'included' }), at({ basis: 'list', usd: 0.03 })]))
- .toEqual({ basis: 'list', usd: 0.05 });
- // One unknown estimate makes the whole set's cost unknown.
- expect(probeTotal([at({ basis: 'list', usd: 0.02 }), at(null)])).toBeNull();
- // A check run only on subscriptions adds no charge.
- expect(probeTotal([at({ basis: 'included' })])).toEqual({ basis: 'included' });
+ expect(probeCostSummary([at({ basis: 'list', usd: 0.02 }), at({ basis: 'included' }), at({ basis: 'list', usd: 0.03 })]))
+ .toEqual({ usd: 0.05, priced: 2, included: 1, unpriced: 0 });
+ // An unknown estimate is counted, never added as zero.
+ expect(probeCostSummary([at({ basis: 'list', usd: 0.02 }), at(null)])).toEqual({ usd: 0.02, priced: 1, included: 0, unpriced: 1 });
+ expect(probeCostSummary([at({ basis: 'included' })])).toEqual({ usd: 0, priced: 0, included: 1, unpriced: 0 });
});
- it('MH-HOSTED-UI-013: offers to check every pair with no current answer, and never an image', () => {
+ it('MH-HOSTED-UI-013: offers to check every pair with no current answer, images included', () => {
const verified = evidence({ status: 'verified', basis: 'probe', observed_at: AT });
const status = {
tools: [
@@ -75,7 +74,7 @@ describe('hosted-tool evidence projection', () => {
};
expect(pendingChecks(status).map(({ tool, candidate: c }) => `${tool}:${c.model_id}`))
- .toEqual(['web_search:gpt-6-astra', 'web_search:gpt-6-luna', 'web_search:gpt-6-terra']);
+ .toEqual(['web_search:gpt-6-astra', 'web_search:gpt-6-luna', 'web_search:gpt-6-terra', 'image_generation:gpt-6.1-sol']);
});
it('MH-HOSTED-UI-006: gives a Source model one capability per tool with evidence, in catalog order', () => {
diff --git a/ui/src/components/settings/models/hostedTools.ts b/ui/src/components/settings/models/hostedTools.ts
index 4011ef8e5..e1d3856f4 100644
--- a/ui/src/components/settings/models/hostedTools.ts
+++ b/ui/src/components/settings/models/hostedTools.ts
@@ -3,7 +3,6 @@
// (`hosted-tools-status.schema.json`): the server has already folded staleness
// into `status`, so nothing here re-derives routing.
import type {
- HostedCostEstimate,
HostedEvidence,
HostedToolCandidate,
HostedToolExecutor,
@@ -70,43 +69,34 @@ export const modelCapabilities = (status: HostedToolsStatus, sourceId: string, m
: [];
});
-/** Tools a user can check. An image probe waits until Vibey parses image results, so the server refuses it. */
-export const probeableTool = (tool: HostedToolName): boolean => tool !== 'image_generation';
-
export type ProbeTarget = { tool: HostedToolName; candidate: HostedToolCandidate };
export const probeKey = ({ tool, candidate }: ProbeTarget): string => `${tool}:${candidate.source_id}:${candidate.model_id}`;
/**
- * What the header check offers to run: every checkable pair with no current
- * answer, i.e. no record, or one an engine upgrade or age made stale.
+ * What the header check offers to run: every pair with no current answer, i.e.
+ * no record, or one an engine upgrade or age made stale.
*/
export const pendingChecks = (status: HostedToolsStatus): ProbeTarget[] =>
- status.tools.flatMap((tool) => (probeableTool(tool.name)
- ? tool.candidates
- .filter((candidate) => candidate.call.status === 'unknown' || candidate.call.stale !== null)
- .map((candidate) => ({ tool: tool.name, candidate }))
- : []));
+ status.tools.flatMap((tool) => tool.candidates
+ .filter((candidate) => candidate.call.status === 'unknown' || candidate.call.stale !== null)
+ .map((candidate) => ({ tool: tool.name, candidate })));
-/**
- * The cost of running these probes, as the server estimates each one. Unknown is
- * never zero: one unknown estimate makes the total unknown. A set run entirely on
- * subscriptions is included.
- */
/** Whether a probe's answer became no evidence: inconclusive, transient, or a request that failed (null). */
export const probeUnanswered = (result: HostedToolProbeResult | null): boolean =>
result === null || result.call.outcome === 'inconclusive' || result.call.outcome === 'transient';
-export const probeTotal = (targets: readonly ProbeTarget[]): HostedCostEstimate | null => {
- let usd = 0;
- let listed = false;
- for (const { candidate } of targets) {
+/**
+ * What running these probes costs, as the server estimates each one: the sum of
+ * the priced ones, and how many are included in a subscription or have no price
+ * (an image, say). Unknown is never added as zero; it is counted.
+ */
+export type ProbeCostSummary = { usd: number; priced: number; included: number; unpriced: number };
+
+export const probeCostSummary = (targets: readonly ProbeTarget[]): ProbeCostSummary =>
+ targets.reduce<ProbeCostSummary>((summary, { candidate }) => {
const estimate = candidate.probe_estimate;
- if (!estimate) return null;
- if (estimate.basis === 'list') {
- usd += estimate.usd;
- listed = true;
- }
- }
- return listed ? { basis: 'list', usd } : { basis: 'included' };
-};
+ if (!estimate) return { ...summary, unpriced: summary.unpriced + 1 };
+ if (estimate.basis === 'included') return { ...summary, included: summary.included + 1 };
+ return { ...summary, usd: summary.usd + estimate.usd, priced: summary.priced + 1 };
+ }, { usd: 0, priced: 0, included: 0, unpriced: 0 });
diff --git a/ui/src/components/settings/models/useEstimateLabel.ts b/ui/src/components/settings/models/useEstimateLabel.ts
index 66146a9bf..f8031d552 100644
--- a/ui/src/components/settings/models/useEstimateLabel.ts
+++ b/ui/src/components/settings/models/useEstimateLabel.ts
@@ -1,6 +1,7 @@
import { useTranslation } from 'react-i18next';
-import { formatSmallUsd } from './format';
+import { formatSentences, formatSmallUsd } from './format';
+import type { ProbeCostSummary } from './hostedTools';
import type { HostedCostEstimate } from './types';
/** 「约 $0.02」, 「订阅内含」, or nothing when the estimate is unknown. */
@@ -13,3 +14,25 @@ export const useEstimateLabel = () => {
: t('settings.models.tools.probe.about', { usd: formatSmallUsd(estimate.usd) }) as string;
};
};
+
+/**
+ * A set of probes' cost in one phrase: the priced sum, then how many are included
+ * or have no price, e.g. 「约 $0.04,另 2 项暂无单价」. Only included reads as
+ * included; nothing priced at all reads as unknown. The locale composes it.
+ */
+export const useCostSummaryLabel = () => {
+ const { t, i18n } = useTranslation();
+ const estimateLabel = useEstimateLabel();
+ return ({ usd, priced, included, unpriced }: ProbeCostSummary): string => {
+ if (priced === 0 && included === 0) return t('settings.models.tools.probe.unknown') as string;
+ const base = priced > 0
+ ? estimateLabel({ basis: 'list', usd }) as string
+ : t('settings.models.tools.probe.included') as string;
+ const extras = [
+ ...(priced > 0 && included > 0 ? [t('settings.models.tools.probe.plusIncluded', { count: included }) as string] : []),
+ ...(unpriced > 0 ? [t('settings.models.tools.probe.plusUnpriced', { count: unpriced }) as string] : []),
+ ];
+ return extras.length === 0 ? base : formatSentences([base, ...extras], i18n.language);
+ };
+};
+
diff --git a/ui/src/i18n/en.json b/ui/src/i18n/en.json
index 5ccc6d921..a2dc5ab38 100644
--- a/ui/src/i18n/en.json
+++ b/ui/src/i18n/en.json
@@ -2379,8 +2379,7 @@
"on": "On",
"off": "Off",
"needsCheck": "Off · check a Source first",
- "needsExecutor": "Off · choose an executor first",
- "checkUnavailable": "Off · checks for this tool come in a later update"
+ "needsExecutor": "Off · choose an executor first"
},
"toggle": {
"turnOn": "Turn on {{name}}",
@@ -2448,6 +2447,14 @@
"included": "included in subscription",
"unknown": "cost unknown",
"notRecorded": "the result couldn't be recorded",
+ "plusIncluded_one": "+ {{count}} included",
+ "plusIncluded_other": "+ {{count}} included",
+ "plusUnpriced_one": "+ {{count}} without a price",
+ "plusUnpriced_other": "+ {{count}} without a price",
+ "replay": {
+ "unverified": "Replay not verified · try again",
+ "unsupported": "Replay not supported on this Source"
+ },
"outcome": {
"inconclusive": "Not verified · the model didn't use the tool; try again",
"transient": "Couldn't verify · try again",
diff --git a/ui/src/i18n/zh.json b/ui/src/i18n/zh.json
index 1981b427b..724e0a08c 100644
--- a/ui/src/i18n/zh.json
+++ b/ui/src/i18n/zh.json
@@ -2379,8 +2379,7 @@
"on": "已开启",
"off": "已关闭",
"needsCheck": "已关闭 · 需先检测",
- "needsExecutor": "已关闭 · 需先选择代理执行者",
- "checkUnavailable": "已关闭 · 此工具的检测将在后续更新中提供"
+ "needsExecutor": "已关闭 · 需先选择代理执行者"
},
"toggle": {
"turnOn": "开启{{name}}",
@@ -2448,6 +2447,14 @@
"included": "订阅内含",
"unknown": "费用未知",
"notRecorded": "结果未能记录",
+ "plusIncluded_one": ",另 {{count}} 项订阅内含",
+ "plusIncluded_other": ",另 {{count}} 项订阅内含",
+ "plusUnpriced_one": ",另 {{count}} 项暂无单价",
+ "plusUnpriced_other": ",另 {{count}} 项暂无单价",
+ "replay": {
+ "unverified": "回放未验证 · 请重试",
+ "unsupported": "此供应商不支持回放"
+ },
"outcome": {
"inconclusive": "未能检测 · 模型没有调用工具,请重试",
"transient": "未能检测 · 请重试",