Skip to content

armgit / macro

Public Imported from GitHub
0

Macro is a unified workspace for teams: email, chat, docs, tasks, agents, calls, and CRM — @-linked together with shared AI memory.

12.7 KB
Raw Blame

Search

Search

Sidebar Search button → /app/search with a focused query box. Results (including a Featured Results group) filter live as you type; no Enter needed. Ctrl+K is usually faster for jump-to-entity; / opens workspace search when no editor is focused.

Routines are found by name in a typed Ctrl+K All query, in touch Search's All name results, and in the desktop Search page's Featured Results. They come from the routines list (Quick Access routine bucket), not the search service, and open the routine in the Agents workspace. They stay out of the empty recency list and out of Home, Drive and Tasks searches.

Desktop filter chips use one compact control size for values, match mode, and clear actions. Labels and controls stay vertically centered inside a frame that grows to fit them, rather than clipping taller segments.

On touch devices, the dock's Search button opens a persistent Search… input in /app/search. It starts in All with Command Menu-style Quick Access name search. Categories are All, Email, Agents, Files, Tasks, Projects, Channels, People in a horizontal pill row at the top, below the status-bar safe area; the query input stays in the bottom dock. Each filter row reads and updates its own Search history entry, including while another entry is in front. Projects respects its feature flag. Categories filter entities directly, independently of view tabs, mailbox selections, and saved facets. Commands and creation shortcuts are excluded. Empty queries show recent entities. Typed All and People name searches also find known teammates before their first DM, with or without CRM. Existing one-to-one DMs replace person rows while preserving name/email matches; selecting a person opens their DM, which the first message creates when there isn't one yet. Verify both categories with a teammate who has no conversation or CRM record. Name search here, in Command K, and in the Chat sidebar finds a conversation by its participants in any order, separated by spaces or commas, with name prefixes allowed. A query naming exactly its participants ranks it right after exact names, above a larger group holding the same people. In Command K and mobile name search, a person's first name alone also counts as exact for their DM or person row. On phones, name-result icons and avatars are scaled to 150% (16px artwork becomes 24px), preserving row height and text alignment.

For a nonempty query, Search full text for “…” appears above name results, including when there are no matches. It retains the query and category and stays in full-text mode while editing or switching categories. Clearing the query, including whitespace-only input, returns to name search and retains the category. Projects is always name-only: selecting it returns to name results and omits the full-text action. People full-text search uses the existing channel-content scope. Name mode makes no full-text requests; explicit full-text mode retains the search service's validation for short queries.

Email Quick Access searches cached thread subjects, sender display names, and sender addresses, including the thread's earlier senders after someone else replies. It returns one row per thread and never matches only a snippet or body. Coverage is existing cached mail: automatic backfill includes 90 days of signal mail and 30 days of noise mail, plus other cached threads. Use full-text search for uncached older mail. No new backfill or server name-search fallback is added. Email name results require the existing normalized GraphQL cache rollout. When verifying a local iOS build without PostHog configuration, enable that path with the build-time VITE_ENABLE_GRAPHQL_SOUP=true override; see the cache guide.

Category/mode changes keep the same input mounted and focused. Return inserts a newline and keeps the keyboard open; the Search bar expands with the text. When enable-universal-input is on, 34 characters (regardless of word count) reveal Universal Input's suggested form above the bar. The editor stays mounted while classification and extraction update the form. Shorter edits keep the form until the input is cleared. While the card is active, the normal search results and top category tabs are hidden; clearing the input restores them. The card stays above the keyboard with its type selector and action buttons visible; scroll the form contents inside the card. Use the explicit action button to create or send, or Save as draft to save an email. Manual field edits keep precedence over later predictions. Event dates and times are edited directly in the card; changing the start preserves the duration. The same PostHog flag as Command K applies; when it is off, the view keeps normal live search. Closing Search clears the input and its unsent suggested form. For Search intent, extraction selects a search type before extracting only that type's supported filters. Task status/priority/assignee/creator filters and channel In/From filters use the same controls and resolution as desktop Command K. The task Assignee facet reuses selected-first ordering when opening the picker, with available agents, team members, and the Search assignees... placeholder. Rows stay in place while selecting or deselecting. Existing selected outsiders remain removable, but contact discovery is not available. Desktop chips and the mobile filter drawer share this behavior. Created by, channel senders, and call speakers retain their existing people pools. The shared global-search filters and results update live above the bar. Open a filtered result directly; there is no separate Search submit action. Opening a result hides search chrome while preserving that Search history entry's query, category, mode, result order, and scroll position. Opening a result does not promote it within that saved list, even when its last-viewed time changes. Swiping back, pressing the native Back control, or using browser Back returns to those results without focusing the input or reopening the keyboard. Name-result ordering and scroll anchors identify a one-to-one DM by its other participant, so a person keeps their place when their first message creates a conversation with a new channel ID. Group conversations keep their own channel identity. Native Search captures lightweight state when covered so deeper navigation can unmount it. History saves the query/category/mode, an account-scoped item ID and offset within that row, a fallback pixel offset, and at most 200 ordering IDs per mode. It never stores result row objects or virtualizer measurement caches. A remount reads current source data; it may show loading while those rows are unavailable. Restoration waits for layout and fetches at most three additional pages to find the anchor. Missing anchors, exhausted/no-progress pagination, or the page limit use the available scroll range instead. Touch, pointer, or wheel interaction cancels pending restoration so later pages cannot move the user. Automatic viewport paging stays paused after recovery until the user interacts or changes the query/category. Matching current rows follow the bounded saved order; absent rows are not revived from history. The live source owns result data and its cache lifetime. Leaving Search through native dock navigation disposes its result view while lightweight entries may remain reachable through history. Starting a new dock Search supplies a fresh route session ID and computes a fresh order. Snapshots are released when their history entry is removed/replaced or its pane closes. Account changes recreate result sources/scrollers and cannot restore another account's anchor or ordering. A failed or empty source cannot revive rows from a historical snapshot. A later search cannot overwrite an earlier entry. Close search ends the session; opening Search from the dock starts in All/name mode. Mobile search state does not modify desktop search state. The dock search row has no Ask AI button. If opening a person fails, search stays open with its query and category intact so the result can be retried.

Verify sender-only teo matches with one- and two-character prefixes, then switch categories with the keyboard up. In both name and full-text modes, scroll several pages, open a result/snippet, and go Back: check the query, category, mode, visible rows, and scroll position. Repeat Back/Forward and native swipe/Back; the keyboard stays dismissed on return. Open Search from the dock and check All/name reset, then revisit the earlier history entry and confirm its query is intact. Open a result from midway through the list and confirm it keeps its original position after Back, both when Search stays mounted and after deeper navigation evicts it. Refresh row metadata and append pages; existing rows must keep their order. Include a person without a DM: open them from midway through All/People results, send them a message, then go Back and confirm that person's new DM is still in the same place and opens the existing conversation on a second selection. Repeat with the DM arriving after Search unmounts, and verify group DMs stay separate. Switch accounts with a scrolled Search entry still mounted, then remount that entry: neither path should retain the previous account's rows or scroll offset. Repeat with a slow refresh: a remount waits for current rows rather than showing saved row objects. Use an anchor on a later page and confirm restoration waits for that page and responsive layout. Remove the anchor or keep it beyond three additional pages and verify recovery stops at the available position. While recovery is pending, scroll manually and confirm later pages do not move the viewport. In full-text mode, repeat Back from a deep list and check the same row returns to the same position when it is available within the recovery limit. Inspect captured Search entry state: it contains bounded IDs and scalar state, with no row objects, snippets, or virtualizer caches. Switch dock views repeatedly and confirm old source rows can be collected even while lightweight history remains. Clear with whitespace and check category retention; select Projects from full-text mode and confirm it shows name results. Repeat opening with a large recent list: results are virtualized, so transferring keyboard focus should not mount every cached row. Scroll through later results to the end to verify additional pages load automatically without a fixed Load more bar. Search results start below the floating category row, with the same clearance on phones and tablets. Check both name and full-text results with the keyboard open and dismissed; categories stay at the top while the input stays above the keyboard. Same-query full-text refreshes keep the current rows visible so paging does not collapse the list and reset its scroll position. Appending a page preserves existing row elements, including when the service rematerializes their data; scroll through multiple pages to check that the position stays put.

Search snippets carry their target in the destination pane's route search. Channel messages open Chat at the message (replies open their parent thread); email snippets open the matching message, Markdown snippets the matching node, PDFs the matching page and highlighted text, agent snippets the matching turn/author, and call snippets the matching transcript segment. Plain rows keep their existing behavior: email rows open normally, while agent and call content results use their first hit.

Verify both a cold open and a result whose entity is already open in another pane, including a Home or Drive detail: reuse keeps that pane's workspace and filters, leaves the search pane intact, and scrolls to the target. Scroll away and click the same snippet again to verify it re-targets. Repeat with Shift-click and Cmd/Ctrl-click (new split and new browser tab), and check Back/Forward restores the earlier target. Channel checks should include an older offscreen message, a reply, and then a root message to ensure the previous thread target is cleared. For a PDF that is still loading, clear its route target before pages become visible. The old search hit must not apply afterward, and its normal initial position should still restore if no target has been applied. A newer mention or preview target must survive that cleanup and still open when the viewer is ready.

Agent-session results use the robot icon and show a highlighted transcript snippet. Show more [N] expands additional matches, labeled User / Agent · Turn N. Click a snippet to open /app/agent/<uuid> at that folded message; a plain row click opens the first content match, or opens the session normally for a title-only hit. These are ACP-backed sessions, distinct from legacy chat results. Legacy chat rename, delete, copy, and move-to-folder actions are not offered on agent-session search rows.