Stability improvements: V0.6 Maybe

[dustinwloring1988]

I have been working on ironing out some of the bugs and errors for my own build but would like to share if you want any.

Summary by CodeRabbit

Release Notes

• Bug Fixes
  • Resolved Linux crash when hiding the dictate pill
  • Enforced 10MB avatar upload limit with allowed extension validation
  • Fixed Personality LLM to use the configured model size
  • Improved CUDA downloads: unsupported platforms fail fast; real download errors stop retries
  • Enhanced long-form Whisper transcription on the PyTorch backend
  • Switched timestamps to timezone-aware UTC across tasks, profiles, stories, and generations
• Documentation
  • Updated GPU acceleration guidance for Linux + NVIDIA
• Tests
  • Added tests covering CUDA download support behavior

[coderabbitai]

📝 Walkthrough

Walkthrough

A focused stability pass applying seven bug fixes: CUDA binary downloads are now gated to Windows only with fast-fail enforcement; personality and compose endpoints respect the configured LLM model; transcription endpoint respects the configured STT model; avatar uploads validate extensions and enforce a 10 MB cap; PyTorch Whisper enables long-form transcription; model generation fails fast when the model is absent; Linux dictate pill crash is prevented via a compile flag; and all datetime.utcnow() calls are replaced with datetime.now(UTC).

Changes

Stability Bug Fixes

Layer / File(s)	Summary
CUDA Windows-only download gating backend/services/cuda.py, backend/routes/cuda.py, app/src/lib/api/types.ts, backend/tests/test_cuda_download.py	New is_cuda_download_supported(), get_cuda_download_unsupported_reason(), and ensure_cuda_download_supported() helpers restrict CUDA binary downloads to Windows. get_cuda_status() gains download_supported and unsupported_reason fields; the download endpoint returns HTTP 409 on unsupported platforms; startup updater early-returns before update checks; CudaStatus TypeScript interface adds the new fields; tests verify platform gating and status shape across Windows and Linux.
Personality and compose endpoints respect configured LLM model backend/routes/generations.py, backend/routes/profiles.py	Both the /generate personality-rewrite path and the profile /compose endpoint now import settings_service, load capture settings, and pass saved.llm_model as model_size to personality.rewrite_as_profile and personality.compose_as_profile respectively.
Transcription endpoint respects configured STT model backend/routes/transcription.py	The /transcribe endpoint now injects a SQLAlchemy session and loads the default speech-to-text model from capture settings when no explicit model is provided, replacing previous non-database model selection logic.
Avatar upload size cap and extension validation backend/routes/profiles.py	Constants AVATAR_MAX_FILE_SIZE (10 MB), AVATAR_UPLOAD_CHUNK_SIZE, and _ALLOWED_IMAGE_EXTS are defined; the upload handler now validates extensions (HTTP 400) and enforces the size cap via chunked reading with running total accounting (HTTP 413).
PyTorch Whisper long-form transcription backend/backends/pytorch_backend.py	Processor call sets truncation=False, padding="longest", return_attention_mask=True; generation passes attention_mask and return_timestamps=True; language hint now uses language and task="transcribe" kwargs instead of forced_decoder_ids.
Model download fast-fail backend/services/generation.py	run_generation() calls tts_model._is_model_cached(model_size) before loading; raises RuntimeError immediately when the model is absent, with graceful fallback if the check is unavailable.
Linux dictate pill crash prevention tauri/src-tauri/src/main.rs	set_ignore_cursor_events(true) in the dictate:hide listener is now wrapped in #[cfg(not(target_os = "linux"))].
Timezone-aware timestamps across models and services backend/database/models.py, backend/services/..., backend/routes/..., backend/utils/tasks.py	All datetime.utcnow() calls across SQLAlchemy model defaults and onupdate handlers, service timestamp writes (profiles, stories, generation, history, export_import, channels), route handlers (profiles, tasks), and task dataclass defaults are replaced with datetime.now(UTC) to use timezone-aware UTC.
Changelog and GPU documentation CHANGELOG.md, docs/content/docs/overview/gpu-acceleration.mdx, docs/content/docs/overview/introduction.mdx	CHANGELOG gains the [Unreleased] section detailing all seven bug fixes; GPU documentation clarifies that Linux+NVIDIA uses local or remote Python backend instead of auto-downloading CUDA, and CUDA backend swap section heading is narrowed to Windows-only.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~25 minutes

Possibly related PRs

• jamiepine/voicebox#252: Introduced the CUDA binary download/restart flow and the original CudaStatus contract that this PR extends with download_supported/unsupported_reason fields.
• jamiepine/voicebox#295: Modified backend/routes/transcription.py to handle Whisper model selection, whereas this PR adds capture settings integration to respect the configured default STT model when none is explicit.

Poem

🐇 Hippity-hop, the bugs hop away,
No more crashes on Linux today!
UTC timestamps, neat and aware,
CUDA stays Windows — that's only fair.
The avatar checks its size at the door,
And Whisper transcribes long speeches once more! 🎤

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 inconclusive)

Check name	Status	Explanation	Resolution
Title check	❓ Inconclusive	The title 'Stability improvements: V0.6 Maybe' is vague and generic, using non-descriptive terms that don't convey meaningful information about the changeset.	Replace with a specific, descriptive title that captures the main change—e.g., 'Fix stability issues: CUDA platform detection, timezone-aware timestamps, and Whisper transcription' or similar.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	Docstring coverage is 87.50% which is sufficient. The required threshold is 80.00%.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

Actionable comments posted: 3

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (1)

backend/routes/tasks.py (1)

89-96: ⚠️ Potential issue | 🟡 Minor

Normalize parsed timestamps to UTC when tzinfo is missing.

Timestamps created in backend/utils/progress.py use datetime.now().isoformat(), which produces naive datetime strings without timezone information. When these strings are parsed with fromisoformat(), the .replace("Z", "+00:00") has no effect (no "Z" present), and fromisoformat() returns a naive datetime. This causes started_at to be inconsistent—sometimes naive, sometimes aware with UTC—despite the fallback on parse failure.

Proposed fix

             if timestamp_str:
                 try:
                     started_at = datetime.fromisoformat(timestamp_str.replace("Z", "+00:00"))
+                    if started_at.tzinfo is None:
+                        started_at = started_at.replace(tzinfo=UTC)
                 except (ValueError, AttributeError):
                     started_at = datetime.now(UTC)
             else:
                 started_at = datetime.now(UTC)

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@backend/routes/tasks.py` around lines 89 - 96, The parsed timestamp from
fromisoformat() may return a naive datetime (without timezone info) since the
original timestamp string lacks a "Z" suffix, making the replace operation
ineffective. After successfully parsing the timestamp_str with fromisoformat()
in the try block, check if the resulting started_at object has no timezone
information (tzinfo is None), and if so, explicitly set it to UTC using the
replace method with tzinfo=UTC. This ensures all parsed timestamps are
consistently timezone-aware datetimes in UTC, matching the fallback behavior in
the else clause.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@backend/database/models.py`:
- Around line 47-48: All DateTime column definitions in the models file that
receive timezone-aware UTC datetimes via datetime.now(UTC) must include the
timezone=True parameter to preserve timezone information through the database
round-trip. Update the created_at and updated_at columns at lines 47-48, and
apply the same timezone=True parameter to all other DateTime columns mentioned
(at lines 85, 96-97, 114, 125-126, 141, 155, 166, 257-258, and 281) to ensure
consistent behavior across all timestamp columns and prevent SQLite from
stripping timezone information on storage and retrieval.

In `@backend/routes/profiles.py`:
- Around line 151-153: Remove ".svg" from the _ALLOWED_IMAGE_EXTS set constant
to prevent SVG uploads which can pose security risks due to active content.
Additionally, ensure that the file extension validation logic (likely used
elsewhere around lines 238-243 as mentioned) properly handles edge cases like
extensionless filenames by requiring that a valid extension is present and
matches one of the allowed raster image extensions in the set. The allowed
extensions should be limited to safe raster formats only: ".png", ".jpg",
".jpeg", ".gif", ".webp", and ".bmp".

In `@backend/services/cuda.py`:
- Line 34: The CUDA_DOWNLOAD_UNSUPPORTED_REASON constant message is misleading
for non-Linux platforms like macOS because it specifically mentions "Linux
users" when the actual gate applies to all non-Windows systems. Update this
constant string to use platform-agnostic language that covers all non-Windows
platforms instead of Linux-specific guidance (e.g., replace "Linux users" with
"users on non-Windows platforms" or similar terminology).

---

Outside diff comments:
In `@backend/routes/tasks.py`:
- Around line 89-96: The parsed timestamp from fromisoformat() may return a
naive datetime (without timezone info) since the original timestamp string lacks
a "Z" suffix, making the replace operation ineffective. After successfully
parsing the timestamp_str with fromisoformat() in the try block, check if the
resulting started_at object has no timezone information (tzinfo is None), and if
so, explicitly set it to UTC using the replace method with tzinfo=UTC. This
ensures all parsed timestamps are consistently timezone-aware datetimes in UTC,
matching the fallback behavior in the else clause.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 83e4e74a-e457-4ca9-93cc-9ef933ba0b60

📥 Commits

Reviewing files that changed from the base of the PR and between b35b909 and 28b7aab.

📒 Files selected for processing (20)

• CHANGELOG.md
• app/src/lib/api/types.ts
• backend/backends/pytorch_backend.py
• backend/database/models.py
• backend/routes/cuda.py
• backend/routes/generations.py
• backend/routes/profiles.py
• backend/routes/tasks.py
• backend/services/channels.py
• backend/services/cuda.py
• backend/services/export_import.py
• backend/services/generation.py
• backend/services/history.py
• backend/services/profiles.py
• backend/services/stories.py
• backend/tests/test_cuda_download.py
• backend/utils/tasks.py
• docs/content/docs/overview/gpu-acceleration.mdx
• docs/content/docs/overview/introduction.mdx
• tauri/src-tauri/src/main.rs