psa-systems/lets-chat
Watch
9
0
Fork
You've already forked lets-chat
 0
Code Issues Pull requests Projects Releases 2 Packages Wiki Activity Actions

feat(home): workspace-wide "Catch me up" AI summary on the dashboard (LC-705) #671

Merged
longjacksonle merged 1 commit from feat/LC-705-workspace-catch-up into main 2026-08-13 18:57:17 +02:00
Conversation 0 Commits 1 Files changed 10 +452
longjacksonle commented 2026-08-13 18:51:18 +02:00
Owner
Copy link

What

Adds a one-click "Catch me up" AI recap of everything the viewer missed across their whole workspace, rendered into a card at the top of the Home dashboard. Unlike the existing per-room "Catch me up" (routes::summary), this reads across every accessible room and DM in a single pass.

Why

The Home dashboard (LC-703) shows what you missed as a card grid, but scanning six cards is still work. This turns "what happened while I was away?" into one button: the AI reads recent unread across every channel and DM and returns a short summary + action items.

How

  • New workspace-level AI gate in ai_gate: allowed_workspace / require_llm_workspace / llm_catch_up_available, reusing the LC-702 flag + audience. "everyone" (default) admits any logged-in user; "staff" narrows to the site-admin role. A cross-room surface has no enclave or per-room override to consult, so "privileged" here reduces to site admin, the same floor privileged_in_room lands on for an enclave-less room.
  • Aggregation reuses db::inbox::list_unread, which returns only accessible rooms, so membership is enforced by the query itself (opening the audience never exposes a non-member's messages). It scans the newest 200 unread, groups per channel, and builds one prompt bounded by MAX_PROMPT_CHARS.
  • No cache, matching the per-room summary: each click regenerates. The result renders through the shared SummaryFragment, so the returned Regenerate button re-runs in place.
  • The entry card renders only when AI is available to the viewer AND there is unread to summarize. The POST re-checks the gate server-side, so the card is a UI affordance, not the access control. If the last unread was read between render and click, the handler returns a friendly caught-up fragment (200), not an error.

Resolved open questions (from the ticket)

  • Cap strategy: newest-200 unread workspace-wide, then trimmed to MAX_PROMPT_CHARS (48k) at prompt-build time.
  • Cache: none. Each click regenerates, same posture as the per-room and transcript summaries.

UI

  • New i18n keys home-catchup-ai-heading / -scope / -caught-up (en + es). Reuses the existing summary-* / ai-* strings for the Generate/Regenerate buttons, disclaimer, and the staged "reading..." pending affordance (driven by the global live.js handler, so it works on the dashboard).

Testing

  • just test: 197 binaries ok, 0 failed, 1664 tests passed.
  • 2 unit tests for the per-channel prompt builder (grouping, chronological ordering, DM labeling, empty-body skipping).
  • 4 route tests for the gate: flag off -> 403 even for an admin; everyone -> a plain member is summarized (200); staff -> member 403 while admin passes; empty workspace -> caught-up fragment (200).
  • cargo clippy clean under standalone and saas; cargo fmt --check clean; i18n en/es parity test passes.

Follow-on to

LC-704 (per-type preview icons) and LC-703 (Home dashboard redesign). This was split out of LC-704 because it is a new cross-room feature rather than a preview tweak.

🤖 Generated with Claude Code

https://claude.ai/code/session_016ufqbuN7mjyt4ms5NEygMy

## What Adds a one-click "Catch me up" AI recap of everything the viewer missed across their whole workspace, rendered into a card at the top of the Home dashboard. Unlike the existing per-room "Catch me up" (`routes::summary`), this reads across every accessible room and DM in a single pass. ## Why The Home dashboard (LC-703) shows *what* you missed as a card grid, but scanning six cards is still work. This turns "what happened while I was away?" into one button: the AI reads recent unread across every channel and DM and returns a short summary + action items. ## How - New workspace-level AI gate in `ai_gate`: `allowed_workspace` / `require_llm_workspace` / `llm_catch_up_available`, reusing the LC-702 flag + audience. `"everyone"` (default) admits any logged-in user; `"staff"` narrows to the site-admin role. A cross-room surface has no enclave or per-room override to consult, so "privileged" here reduces to site admin, the same floor `privileged_in_room` lands on for an enclave-less room. - Aggregation reuses `db::inbox::list_unread`, which returns only accessible rooms, so membership is enforced by the query itself (opening the audience never exposes a non-member's messages). It scans the newest 200 unread, groups per channel, and builds one prompt bounded by `MAX_PROMPT_CHARS`. - No cache, matching the per-room summary: each click regenerates. The result renders through the shared `SummaryFragment`, so the returned Regenerate button re-runs in place. - The entry card renders only when AI is available to the viewer AND there is unread to summarize. The POST re-checks the gate server-side, so the card is a UI affordance, not the access control. If the last unread was read between render and click, the handler returns a friendly caught-up fragment (200), not an error. ## Resolved open questions (from the ticket) - Cap strategy: newest-200 unread workspace-wide, then trimmed to `MAX_PROMPT_CHARS` (48k) at prompt-build time. - Cache: none. Each click regenerates, same posture as the per-room and transcript summaries. ## UI - New i18n keys `home-catchup-ai-heading` / `-scope` / `-caught-up` (en + es). Reuses the existing `summary-*` / `ai-*` strings for the Generate/Regenerate buttons, disclaimer, and the staged "reading..." pending affordance (driven by the global `live.js` handler, so it works on the dashboard). ## Testing - `just test`: 197 binaries ok, 0 failed, 1664 tests passed. - 2 unit tests for the per-channel prompt builder (grouping, chronological ordering, DM labeling, empty-body skipping). - 4 route tests for the gate: flag off -> 403 even for an admin; `everyone` -> a plain member is summarized (200); `staff` -> member 403 while admin passes; empty workspace -> caught-up fragment (200). - `cargo clippy` clean under `standalone` and `saas`; `cargo fmt --check` clean; i18n en/es parity test passes. ## Follow-on to LC-704 (per-type preview icons) and LC-703 (Home dashboard redesign). This was split out of LC-704 because it is a new cross-room feature rather than a preview tweak. 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_016ufqbuN7mjyt4ms5NEygMy
feat(home): workspace-wide "Catch me up" AI summary on the dashboard (LC-705)
All checks were successful
check-secrets / Nosey parker (push) Successful in 4s
Details
check-secrets / TruffleHog (push) Successful in 5s
Details
check-secrets / Kingfisher (push) Successful in 7s
Details
check-secrets / Nosey parker (pull_request) Successful in 4s
Details
check-secrets / TruffleHog (pull_request) Successful in 4s
Details
check-secrets / Kingfisher (pull_request) Successful in 12s
Details
Check / clippy + fmt + tests (pull_request) Successful in 5m57s
Details
Create release / Create release from merged PR (pull_request) Has been skipped
Details
752c03fe40 
Adds a one-click AI recap of everything the viewer missed across every accessible room and DM, rendered into a card at the top of the Home dashboard. Unlike the existing per-room "Catch me up" (routes::summary), this is cross-room: it gathers recent unread workspace-wide, groups it per channel, and summarizes in a single pass.

Gate: a new workspace-level AI gate (ai_gate::allowed_workspace / require_llm_workspace / llm_catch_up_available) reuses the LC-702 flag + audience. "everyone" (default) admits any logged-in user; "staff" narrows to the site-admin role (a cross-room surface has no enclave or per-room override to consult, so "privileged" reduces to site admin, the same floor privileged_in_room lands on for an enclave-less room). The POST re-checks the gate server-side, so the card is a UI affordance, not the access control.

Aggregation reuses db::inbox::list_unread (accessible rooms only, so membership is enforced by the query), scans the newest 200 unread, and builds one per-channel-grouped prompt bounded by MAX_PROMPT_CHARS. No cache, matching the per-room summary: each click regenerates. The result renders through the shared SummaryFragment, so the returned Regenerate button re-runs in place. The entry card shows only when AI is available to the viewer AND there is unread to summarize; the handler still returns a friendly caught-up fragment (not an error) if the last unread was read between render and click.

New i18n keys home-catchup-ai-heading / -scope / -caught-up in en and es; reuses the existing summary-* / ai-* strings for the button, disclaimer, and pending affordance.

Tests: 2 unit tests for the per-channel prompt builder (grouping, chronological ordering, DM labeling, empty-body skipping) and 4 route tests for the gate (flag off -> 403 even for admin; everyone -> member summarized; staff -> member 403 / admin 200; empty workspace -> caught-up 200).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016ufqbuN7mjyt4ms5NEygMy
longjacksonle scheduled this pull request to auto merge when all checks succeed 2026-08-13 18:52:21 +02:00
longjacksonle deleted branch feat/LC-705-workspace-catch-up 2026-08-13 18:57:18 +02:00
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
psa-systems/lets-chat!671
No description provided.