Research runPro
A deep-research job: phases, the queries it ran, sources rolling in, progress and elapsed time, ending on a report.
oxyn add research-run- Planning
- Searching
- Reading
- Writing
The preview stays free for everyone. Pro opens the code of every Pro component, screen, agent-app template, block and kit, plus the Pro CLI.
Rendered from Rust at build time · goes live when visible
When to use
- An agent answer is only as useful as the user's ability to check it, trust it, and continue from it.
- Inline numbered citations with hover previews and a sources strip are now common; Perplexity loads source cards before the answer streams (note 04); follow-up / related chips and starter grids are everywhere.
- The open gaps our research found are exactly where an agent kit should lead: memory transparency — only Le Chat shows "clickable receipts" (when, where from, why), ChatGPT shows "Memory updated", nobody shows a memory diff or "memory used" on an answer — and claim provenance (sourced vs inferred vs guessed), seen only in the assistant-ui catalogue.
- This spec makes both first-class, with the same state discipline as the rest of Oxyn.
When not to use
- Markdown that renders correctly while it streams: use Streaming markdown instead.
- Streamed text arriving: use Text Reveal instead.
- Feedback on a reply: use Response feedback instead.
Installation
Research run is part of Oxyn Pro; the preview is free. Pro items install with the oxyn CLI once your licence is signed in; until then Pro members copy the code from this page.
oxyn add research-runUsage
The preview is free. Pro opens the code of every Pro component, screen and template.
Examples
Running
Planning, searching, reading, writing.
- Planning
- Searching
- Reading
- Writing
Code Pro
The preview is free. Pro opens the code of every Pro component, screen and template.
Done
The finished report card.
- Planning
- Searching
- Reading
- Writing
Code Pro
The preview is free. Pro opens the code of every Pro component, screen and template.
API reference
ResearchRun
| Prop | Type | Default | Description |
|---|---|---|---|
questionrequired | String | — | Question. |
queriesrequired | Vec<String> | — | Queries. |
sourcesrequired | Vec<(String, String)> | — | Sources (title, domain). |
elapsed_ms | u64 | u64::MAX | ms since start. |
Read from the Rust signature and doc comments.
Keyboard interactions
| Key | Action |
|---|---|
| Tab | Pause, Stop, Open report |
Accessibility
- A progressbar with a value; the current phase has aria-current=step.
From the spec
- Citation marker: a link-styled button, name "Citation 1: nature.com, ‘Title…’" (domain + title when resolved);
aria-describedbythe preview when open; preview is a non-modal dialog (role="dialog", labelled by the source title) reachable with Tab from the marker and closed with Esc; focus returns to the marker. Hover-only content is also available on focus (WCAG 1.4.13: dismissible, hoverable, persistent). - Sources list:
role="list"; cards are links with "Source 3 of 10, cited" / "not cited"; filter is a radio group. - Searching: status updates via the composer live region at most every 10 s ("Found 6 sources").
- Confidence: badge text is the accessible name ("Inferred claim"); glyph
aria-hidden; colour never carries meaning alone. - Memory receipts: "Saved to memory: prefers metric units. Undo available for 10 seconds" announced politely once; Undo is a button; the 10 s window extends while the receipt has focus (WCAG 2.2.1 — proposed: timer pauses on focus/hover).
- Follow-ups:
role="group"labelled "Suggested follow-ups"; each chip a button; not announced on appearance (they are optional), but the group is the next Tab stop after the last message. - Empty state: greeting is a heading; starters are buttons in a group "Suggestions"; the composer keeps initial focus.
- Contrast: markers ≥ 4.5:1 (they are text); dimmed uncited cards ≥ 4.5:1 for text; skeletons decorative (
aria-hidden). - Reduced motion: §6 last row.
Motion & interaction
- Marker resolve: Pending → Resolved opacity 60 → 100 % (120 ms).
- Preview open: after
delay.preview.open300 ms; fade + scale 0.98 → 1 + y 4 → 0 (160 msease.out); the cited span inside the snippet gets a highlight sweep left → right (240 ms, once). Close afterdelay.preview.close150 ms grace (moving the pointer from marker to card keeps it open). Moving between adjacent markers swaps content in place (120 ms crossfade, no close/open). - Marker ↔ card link: hovering a marker rings its card in the list/rail; hovering a card underlines its markers in the answer (both 120 ms).
- Searching: cards fill one by one as results arrive:
dur.row.enter180 ms, stagger 40 ms, cap 5 (stack variant: favicons slide into the stack with −6 px overlap, 160 ms). - Stack → list expand: favicons spread into rows (200 ms
spring.snap). - Re-order on done: cards glide to cited-first order (220 ms
spring.snap). - Confidence tooltip: 300 ms delay, 120 ms fade.
- Memory saved: chip enters with
dur.row.enter; memory glyph does one 400 ms pulse (scale 1 → 1.15 → 1); Undo underline shrinks over 10 s (linear); Undone crossfade 160 ms then exit 200 ms. - Memory updated: before → after shown as strike-through fading out (300 ms) while new text fades in.
- Follow-ups: appear 300 ms after Done; chips rise 6 px + fade (180 ms), stagger 40 ms, cap 5; on choose, the chip text flies to the composer (fill: 220 ms
ease.in-out) or to the transcript (send: ≤ 260 ms, same as composer fly-up); the other chips fade out 120 ms. - Empty state: starters stagger in on mount (180 ms, 40 ms, cap 4); on first keystroke they fade (160 ms) and the area collapses (200 ms); on send the hero composer docks (Pro
glass-hero,dur.travel260 msease.in-out).
Timing
| Transition | Duration | Delay | Easing / spring | Stagger | Frame rate |
|---|---|---|---|---|---|
| Marker resolve | 120 ms | 0 | ease.out | — | 60 Hz |
| Preview open / close | 160 / 120 ms | 300 ms / 150 ms grace | ease.out / ease.in | — | 60 Hz |
| Snippet highlight sweep | 240 ms once | on open | ease.out | — | 60 Hz |
| Preview content swap (adjacent markers) | 120 ms | 0 | ease.in-out | — | 60 Hz |
| Marker ↔ card ring / underline | 120 ms | 0 | ease.out | — | 60 Hz |
| Source card enter | 180 ms, y +4 → 0 | skeletons after 150 ms | ease.out | 40 ms, cap 5 | 60 Hz |
| Stack → list | 200 ms | 0 | spring.snap | — | 60 Hz |
| Re-order on done | 220 ms | 0 (held while hovered) | spring.snap | — | 60 Hz |
| Card highlight flash | 600 ms hold + 900 ms fade | 0 | ease.out | — | 30 Hz |
| Confidence tooltip | 120 ms | 300 ms | ease.out | — | 60 Hz |
| Memory chip enter / glyph pulse | 180 / 400 ms | 0 | ease.out / spring.soft | — | 60 Hz |
| Undo window | 10 000 ms underline | 0 | linear | — | 15 Hz |
| Updated before → after | 300 ms | 0 | ease.in-out | — | 60 Hz |
| Follow-ups enter | 180 ms, y +6 → 0 | 300 ms after Done | ease.out | 40 ms, cap 5 | 60 Hz |
| Follow-up chosen (fill / send) | 220 / ≤ 260 ms | 0 | ease.in-out / ease.in | — | 60 Hz |
| Starters enter / fade on typing / collapse | 180 / 160 / 200 ms | 0 | ease.out / ease.in / spring.snap | 40 ms, cap 4 | 60 Hz |
| Hero composer dock (Pro) | 260 ms | on first send | ease.in-out | — | 60 Hz |
| Reduced motion | all → 0–80 ms opacity crossfades; no sweeps, pulses, glides or staggers; previews still wait 300 ms (intent detection, not motion); Undo shown as "Undo (10 s)" text counting at 1 Hz; follow-ups and starters appear together | — | linear | none | on change |
Responsive behaviour
- Columns stack below 520 px of container.
Edge cases (spec)
- Many citations (50+): markers stay inline; the list collapses to the stack with "50 sources"; numbers ≥ 10 widen the pill to the 2-digit reserved width (status-text
ReserveMagnitude). - Citations streaming before sources resolve:
Pendingmarkers; if a marker never resolves byDone, it becomesOrphan(no broken link). - Duplicate sources: deduplicated by canonical URL; markers point to the single card.
- Long titles / URLs: titles clamp at 2 lines; domains middle-truncate; full in the tooltip.
- RTL: markers follow the claim at the logical end; numbers in bidi isolates; cards row scrolls from the right; favicon stack overlaps toward the inline-end.
- Tiny sizes:
citation/numberat 10 px min font with a 24 px hit area (IDE) / 44 px (touch) via an invisible padded target;confidence-badge/iconfor ≤ 16 px rows. - Dense IDE:
source-list/listcollapsed by default to "6 sources ›". - Touch: previews are tap-to-pin sheets; follow-up chips scroll horizontally when they would wrap past 2 rows.
- Slow networks: source cards keep skeletons; favicons fall back to a monogram tile after 1 s.
- Very fast agents: sources and answer land together — cards render settled, no fill animation.
- Very slow searches:
Searching(n)count and the status line carry liveness; skeletons stop sweeping afterstall_softand show "Still searching…". - Memory receipt on a branch: receipts belong to the version that created them; switching versions shows that version's receipts; Undo on an old version asks for confirmation (the memory may be used since).
- Follow-ups after an error or interrupt: not shown (proposed); the message's Retry is the next step.
- Empty state with a pre-filled draft (deep link): starters hidden, composer focused.
- Starter text longer than the composer limit: rejected at authoring (codegen check, 2 000 chars).
Notes for AI
- Import
ResearchRunfromoxyn_dioxusand includeSTATUS_INDICATOR_CSSplus the token stylesheet once. - There is no GPUI version yet; use the Dioxus component.
- Props to set first:
question,queries,sources. - It reads a clock (
now) for live timers; pass one clock for the whole page rather than one per component. - Keep its states and timings as they ship; restyle it through the CSS variables rather than by hand.