Summary generation fails with error icon at end of meeting

[hyper-charlie]

Description

After a meeting ends, the app attempts to generate a summary but shows only an error icon with no explanation message. The error icon appears after a short delay.

Steps to Reproduce

1. Complete a meeting
2. Wait for the app to automatically attempt summary generation
3. Observe error icon appears instead of summary

Expected Behavior

Meeting summary should generate and display without errors, or provide a clear error message if generation fails.

Actual Behavior

Error icon appears with no explanation text. Summary is never shown to the user.

Environment

• Platform: macOS
• OS: 26.2.0
• App Version: 1.0.6
• Build: 9820787
• Locale: en-US

Additional Notes

• Error appears after short delay following meeting end
• User ID: 1da7d2c3-dea4-4cd7-aa71-972985bf1207

[hyper-charlie]

Additional Context

User reports this occurs on all meetings, not just isolated cases. Summary generation consistently fails with error icon appearing after meeting end.

[devin-ai-integration]

Codebase exploration for summary generation error

I traced the full code path from meeting end → auto-enhance trigger → error display. Here's what I found:

Relevant Files

File	Role
apps/desktop/src/hooks/useAutoEnhance.ts	Orchestrates auto-enhance after meeting ends
apps/desktop/src/hooks/autoEnhance/trigger.ts	Detects listener stop (active → inactive)
apps/desktop/src/hooks/autoEnhance/runner.ts	Runs the enhance task, creates enhanced note
apps/desktop/src/hooks/autoEnhance/eligibility.ts	Checks min word count (5 words)
apps/desktop/src/store/zustand/ai-task/tasks.ts	Core task execution with error handling (lines 217-249)
apps/desktop/src/store/zustand/ai-task/task-configs/enhance-workflow.ts	Enhance workflow: template generation + summary streaming
apps/desktop/src/store/zustand/ai-task/task-configs/enhance-transform.ts	Gathers session/transcript data for the LLM prompt
apps/desktop/src/components/main/body/sessions/note-input/enhanced/index.tsx	Renders enhanced note view (or nothing on error)
apps/desktop/src/components/main/body/sessions/note-input/header.tsx	Tab header with error icon (lines 289-317) and useEnhanceLogic (lines 680-771)
apps/desktop/src/hooks/useLLMConnection.ts	Resolves LLM provider/model connection

Root cause: error state renders icon but no message

1. Error icon shows in tab header, but content area is blank

In enhanced/index.tsx (line 39-41):

if (status === "error") {
  return null;  // ← renders nothing in the content area
}

Meanwhile in header.tsx (lines 289-306), HeaderTabEnhanced shows AlertCircleIcon when isError is true — but there's no tooltip or message explaining the error. So the user sees a red error icon in the tab with a blank content area and zero context.

2. Error object is captured but never displayed

In tasks.ts (lines 234-247), when the enhance task fails, the error is stored:

const error = extractUnderlyingError(err);
// sets status: "error" and error object in zustand state

The useEnhanceLogic hook (header.tsx:759) reads this error, but it's never rendered anywhere — not in the tab header, not in the content area, not as a toast.

Likely failure causes for this user

Given the user has app version 1.0.6 and the error appears "after a short delay" (consistent with an LLM API call failing), probable causes are:

1. LLM API error — The most likely scenario. The enhance-workflow.ts calls streamText() / generateText() which can fail due to auth issues, rate limits, invalid API key, or provider downtime. Errors from the AI SDK are caught in tasks.ts:217 and stored, but the extractUnderlyingError function (line 253) tries to unwrap AI_RetryError — if the underlying cause is an API auth error or network timeout, it gets swallowed into a generic error object.

2. Template generation failure — generateTemplateIfNeeded() (enhance-workflow.ts:127) calls generateText() with structured output. If the model returns unparseable JSON, the fallback at line 215-221 tries regex extraction. Both failures return null silently — but the workflow continues, which may lead to a bad prompt → LLM error.

3. Validation exhaustion — withEarlyValidationRetry (validate.ts:73) throws after maxRetries (2) if the LLM output doesn't start with a markdown heading. This becomes the error the user sees.

4. no_model case is silently dropped — In useAutoEnhance.ts (line 33-36), when runner.run() returns { type: "no_model" }, it's not handled — only "skipped" is checked. The note is never created but the tab might already be switched.

Recent changes in this area

• 931a5f5fb — "feat(desktop): AI-powered note enhancement with auto-trigger and batch improvements (feat(desktop): AI-powered note enhancement with auto-trigger and batch improvements #3903)" (today) — Modified autoEnhance/runner.ts, useAutoEnhance.ts, and batch/listener state. This is the most recent change to the auto-enhance flow and may be relevant.
• 5dae51ff4 — "fix: persist summary via onComplete to survive tab unmount (fix: persist summary when tab is not active during generation #3881)" — Added onComplete callback pattern to persist summary even when the tab unmounts during generation.

Suggested investigation areas

1. Display the error message to the user — The enhanced/index.tsx error branch (line 39) returns null. It should instead show the error message from taskState.error, similar to how ConfigError handles missing config. The error text is already available via useAITaskTask.

2. Add error context to the tab icon — The AlertCircleIcon in header.tsx:302 has no tooltip. Adding a title/tooltip with error?.message would give users immediate context.

3. Handle no_model in useAutoEnhance — The runner.run() return type "no_model" is not displayed as a skip reason, so the user gets no feedback when no LLM is configured.

4. Check if this is reproducible with a specific LLM provider — The user's environment info doesn't include which provider/model they're using. Knowing whether they use hyprnote (hosted) or a third-party provider would help narrow down whether this is an API-side or client-side issue.