Summary generation displays [object Object] error on all notes

[hyper-charlie]

Description

Summary generation fails with [object Object] error message displayed instead of the actual summary. This occurs consistently across all notes.

Steps to Reproduce

1. Record or create a meeting note (tested with 35-minute recording)
2. Attempt to generate a summary
3. Error message appears instead of summary

Expected Behavior

Summary should generate and display the AI-enhanced summary text

Actual Behavior

[object Object] error is displayed

Environment

• Platform: macOS
• Architecture: aarch64
• OS Version: macOS 26.2
• App Version: 1.0.11
• Build: 0b5cb58
• Locale: en-US

Impact

User unable to generate summaries for any notes, affecting core app functionality

[hyper-charlie]

Additional Details

• Retry button present but non-functional - Clicking Retry produces no change
• Model Provider: OpenAI
• Internet Connection: Stable
• Scope: Affects all notes consistently

The error appears to be in response handling rather than connectivity or configuration. The app is receiving data from OpenAI but failing to parse/display it properly.

[devin-ai-integration]

Investigation Findings

I traced the full error flow for summary generation and identified the most likely root cause of the [object Object] display.

Error Display Path

The error message is rendered in two places:

1. enhance-error.tsx:44 — {error?.message || "Something went wrong while generating the summary."}
2. header.tsx:130 — error.message shown in the tab header

Both rely on error being a proper Error instance with a .message string. If error.message is itself an object (or error is not a real Error), React will render it as [object Object].

Root Cause Analysis

There are two likely paths that could produce [object Object]:

Path 1: extractUnderlyingError in tasks.ts (lines 280-311) — falling through to return err

When the AI SDK throws a RetryError (name: "AI_RetryError"), the function tries to extract the underlying cause. However:

• RetryError.lastError and RetryError.errors[] are typed as unknown in AI SDK v6. If the underlying provider error is a plain object (e.g., a JSON response body like { error: { message: "...", type: "..." } }) rather than an Error instance, all the instanceof Error checks fail.
• The function falls through to return err (line 310), returning the RetryError itself. The RetryError.message is something like "Failed after 2 attempts. Last error: [object Object]" because getErrorMessage() in the AI SDK calls String() on non-Error objects.

So error.message would literally contain [object Object].

Path 2: Template render failures via Tauri plugin bindings

In enhance-workflow.ts, lines 77/109/126 do:

throw new Error(result.error);

The result.error comes from the template plugin's auto-generated bindings (plugins/template/js/bindings.gen.ts:14):

else return { status: "error", error: e as any };

If the Tauri invoke throws a non-string error (a serialized Rust error object), then new Error(someObject) produces an Error with message [object Object].

However, the Rust side (plugins/template/src/commands.rs:9) does .map_err(|e| e.to_string()), so this path should produce string errors. Path 1 is more likely.

Most Likely Scenario

The user's LLM provider returns an API error (rate limit, auth failure, context length exceeded, etc.) that gets wrapped as a non-Error object inside the AI SDK's retry mechanism. The extractUnderlyingError function doesn't handle this case, resulting in [object Object] in the error message.

Relevant Files

File	Role
apps/desktop/src/store/zustand/ai-task/tasks.ts	Core task execution + extractUnderlyingError (lines 280-311)
apps/desktop/src/store/zustand/ai-task/task-configs/enhance-workflow.ts	Summary generation workflow
apps/desktop/src/session/components/note-input/enhanced/enhance-error.tsx	Error display component
apps/desktop/src/session/components/note-input/enhanced/index.tsx	Routes to error/streaming/editor views
apps/desktop/src/services/enhancer/index.ts	Enhancer service orchestration
apps/desktop/src/ai/hooks/useAITaskTask.ts	Hook that exposes task state (including error) to components

Recent Related Changes

• eb2b135 — AI SDK v5→v6 migration (changed generateObject → generateText with Output.object). This changed how errors are structured.
• 9918d36 — Added fallback for models without tool calling support
• 344fa6c — Fixed graceful yield when validation exhausts retries
• b9da674 — Pre-meeting note support for better summaries

Suggested Fix

In extractUnderlyingError (tasks.ts), handle the case where lastError/errors[] are plain objects:

function extractUnderlyingError(err: unknown): Error {
  if (!(err instanceof Error)) {
    return new Error(typeof err === 'object' ? JSON.stringify(err) : String(err));
  }

  if (err.name === "AI_RetryError") {
    if ("cause" in err && err.cause instanceof Error) {
      return err.cause;
    }

    if ("lastError" in err) {
      const lastError = (err as any).lastError;
      if (lastError instanceof Error) return lastError;
      // Handle non-Error lastError (plain objects from provider)
      if (lastError != null) {
        return new Error(typeof lastError === 'object' ? JSON.stringify(lastError) : String(lastError));
      }
    }

    // ... rest of existing logic
  }

  return err;
}

Additionally, consider adding a safety net in enhance-error.tsx to guard against non-string messages:

const message = typeof error?.message === 'string' 
  ? error.message 
  : "Something went wrong while generating the summary.";

Additional Context

• The issue is consistent across all notes because it's triggered by the LLM provider response, not note content.
• The user's build hash 0b5cb5841b9dd0a28f4e4585a42cb3ea86828959 and app version 1.0.11 would help pinpoint which exact code they're running.
• It would be helpful to know which LLM provider/model the user has configured, as different providers return different error shapes.