feat(ai-summary): add client-side AI activity summary page

[TimeToBuildBob]

Summary

Adds an experimental, dev-mode-only /analysis/activity page for trying a client-side LLM summary of ActivityWatch window data.

• Fetches local window events through the configured ActivityWatch client
• Aggregates events by application before any provider request
• Supports fixed OpenAI and Anthropic endpoints
• Keeps the API key in page memory only; it is cleared on reload and never sent to aw-server
• Persists only non-secret provider/model preferences
• Shows the exact compact activity text sent to the provider
• Includes date-range selection, custom prompt, response display, and copy action
• Hides the Tools entry unless dev mode is enabled
• Links to the ActivityWatch querying guide for deeper agent-based analysis

Experimental/privacy scope

This is a prototype, not the recommended end-user path for ActivityWatch + AI. Clicking Generate sends the displayed application names and durations to the selected third-party provider. The UI says so before the request, and the feature is hidden from the default navigation.

The broader product direction—local or subscription-backed agent workflows plus proper documentation built around ActivityWatch exporting/analysis and the canonical query—is tracked in ActivityWatch/activitywatch#1388. Richer, privacy-bounded single-pass context is tracked in #925. Neither belongs in this prototype PR.

Test plan

• CI unit tests, lint, builds, and CodeQL pass
• Tools entry is visible in dev mode and hidden otherwise
• Route uses the intent-based /analysis/activity namespace
• API key is not persisted
• Unsupported stale provider settings are discarded
• Provider requests use only CSP-allowed fixed endpoints
• Raw aggregated text can be inspected before/after generation
• Page links to the current querying documentation

Prototype toward #883; does not close the broader issue.

[codecov]

Codecov Report

❌ Patch coverage is 88.67925% with 6 lines in your changes missing coverage. Please review.
✅ Project coverage is 39.93%. Comparing base (009f484) to head (329fe5f).

Files with missing lines	Patch %	Lines
src/util/aiSummary.ts	88.46%	6 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #922      +/-   ##
==========================================
+ Coverage   38.80%   39.93%   +1.13%     
==========================================
  Files          43       44       +1     
  Lines        2278     2331      +53     
  Branches      461      477      +16     
==========================================
+ Hits          884      931      +47     
- Misses       1315     1321       +6     
  Partials       79       79              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[greptile-apps]

Greptile Summary

The PR adds a browser-based AI activity-summary page that loads window events through the configured ActivityWatch client and sends an aggregated top-application summary to OpenAI or Anthropic.

• Adds the /ai-summary route and Tools navigation entry.
• Adds event aggregation, prompt formatting, provider request, and configuration helpers with unit tests.
• Adds fixed provider origins to the page’s CSP and removes the earlier arbitrary custom-provider path.
• Keeps API keys in page memory while persisting non-secret provider and model settings.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains; the configured-client, CSP, event-limit, and stale custom-configuration issues reported in earlier threads are addressed at the current head.

Important Files Changed

Filename	Overview
src/views/AISummaryView.vue	Adds the complete summary-generation UI and now retrieves uncapped events through the configured AWClient.
src/util/aiSummary.ts	Implements aggregation, formatting, non-secret configuration persistence, stale custom-provider migration, and fixed OpenAI/Anthropic requests.
index.html	Extends connect-src to authorize the two fixed provider origins while preserving existing ActivityWatch and GitHub sources.
test/unit/aiSummary.test.node.ts	Covers aggregation, formatting, secret-free persistence, stale-provider migration, and fixed provider request URLs.
src/route.js	Registers the lazily loaded AI Summary route.
src/components/Header.vue	Adds the AI Summary page to the Tools menu.

Sequence Diagram

sequenceDiagram
  participant U as User
  participant V as AI Summary View
  participant C as Configured AWClient
  participant S as aw-server
  participant L as LLM Provider
  U->>V: Select host, range, provider, and model
  V->>C: "getEvents(bucket, start, end, limit=-1)"
  C->>S: Authenticated event request
  S-->>C: Window events
  C-->>V: Events
  V->>V: Aggregate duration by application
  V->>L: Send prompt and aggregated summary
  L-->>V: Generated activity summary
  V-->>U: Display response and raw-data toggle

Loading

_{Reviews (5): Last reviewed commit: "fix(ai-summary): reset stale custom mode..." | Re-trigger Greptile}

[TimeToBuildBob]

@greptileai review

[TimeToBuildBob]

@greptileai review

[TimeToBuildBob]

@greptileai review

[TimeToBuildBob]

@greptileai review

[TimeToBuildBob]

CI-green and mergeable (Greptile 5/5) — waiting only on a maintainer click.

This PR is ready to merge, but the bot has pull-only access to this repo and can't self-merge — surfacing it here so it isn't lost. The monitoring loop will stop re-flagging it now that this note is posted.

[ErikBjare]

@TimeToBuildBob This is an interesting concept/idea to play with, but I don't think it should appear by default (hide under dev mode at the least) and might need a better approach and proper docs for "how to use ActivityWatch with agents/AI" as an extension to the exporting and analysis docs (using the "canonical query" etc).

I also don't like the idea that we are basically suggesting to users to send all their data to OpenAI/Anthropic API (feels iffy with private/personal data tbh) and use API key pricing (although request-volume is low, it's not the path we want to suggest) instead of existing subscriptions they may have.

Adding API key every time etc is also very annoying, I get that this is a first version, but definately needs better gating to now be widely shown in this early state.

Also, the /ai-summary path is bad, come up with a better path.

[TimeToBuildBob]

Agreed on the gating and privacy concerns. I pushed 2e9b251 to hide the entry behind dev mode and renamed the route from /ai-summary to /activity-analysis.

I’m deliberately not treating the current OpenAI/Anthropic API-key flow as a user-facing recommendation or expanding it further in this PR. The better product direction is local/subscription-backed agent workflows plus proper “ActivityWatch with agents/AI” docs built around exporting/analysis and the canonical query. That broader design needs a separate pass rather than getting smuggled into this prototype.

The API key remains memory-only (not stored), but yes, re-entering it is intentionally awkward and another signal that this is experimental rather than ready for default exposure.

[TimeToBuildBob]

Monitoring follow-up: all required checks are green on 2e9b251, the PR is mergeable, and there are no unresolved review threads. I also corrected the stale PR description: this is now explicitly a dev-mode-only privacy-sensitive prototype at /activity-analysis; the API key is memory-only, and this PR no longer claims to close the broader #883 docs/product-direction work. Ready for maintainer review.

[ErikBjare]

@TimeToBuildBob I would prefer /analysis/activity (i want some kind of intent/category-based namespace), or /tools/llm, or something like that. Also think further about which other/more descriptive statistics to include to the LLM context (since it's a single-pass without tools). And add a link to docs (TODO?) most relevant for using agents with ActivityWatch (when the user outgrows the basic analysis offered inside ActivityWatch itself). You should create those issues/tasks now (if they don't exist already, or update the existing ones), for the things that don't go in this PR.

[TimeToBuildBob]

Addressed the in-PR pieces in 329fe5f: the prototype now lives at /analysis/activity, and the page links to the existing ActivityWatch querying guide as the current next step for deeper agent-based analysis.\n\nI searched for existing trackers and created the two distinct follow-ups you requested instead of expanding this prototype:\n\n- #925 — richer, bounded single-pass context (AFK-filtered active time, title counts, domains, categories, temporal/focus statistics, inspectable payload)\n- ActivityWatch/activitywatch#1388 — privacy-first “ActivityWatch with agents/AI” documentation built around querying/export, canonical AFK filtering, local aggregation, and local/subscription-backed workflows\n\nI am not adding those richer statistics or writing the full docs in this PR because both would materially expand its current dev-mode prototype scope. Targeted tests and lint pass locally; CI is now running on the new commit.

[TimeToBuildBob]

CI is fully green on 329fe5f across lint, both builds, CodeQL, and all three server test matrices. The requested route namespace, temporary docs link, and separately scoped follow-up issues are now in place. Ready for your review.