[LUM-753] Improve client-side dictation resilience#24042

Merged

noanflaherty merged 2 commits into

mainfrom

devin/1775585427-LUM-753-dictation-resilience

Apr 7, 2026

Contributor

devin-ai-integration Bot commented Apr 7, 2026 •

edited

Loading

Summary

Tunes the existing client-side dictation fallback to be less blocking when the daemon is unreachable (cloud-hosted assistants). The client already falls back to raw Apple Speech Recognition transcription on HTTP failure — this PR makes that happen faster and with better diagnostics.

Changes:

Reduce dictation HTTP timeout from 10s → 5s so users see their transcription sooner instead of staring at a "Processing…" spinner when the daemon can't be reached
Downgrade failure logs from error → warning since these are expected failures with graceful recovery, not actionable errors
Add elapsed time to all failure log messages for latency debugging
Include transcription length in fallback log for diagnostics

No new fallback logic — the fallbackResponse / fallbackMode mechanism was already in place.

Review & Testing Checklist for Human

Confirm 5s timeout is acceptable. Cloud-hosted assistants route through the platform proxy which can add ~6s of actor-token bootstrap latency. With a 5s client timeout, cloud dictation will consistently fall back to raw transcription until the platform proxy is also fixed (PR #4007, currently draft). Verify this is the intended behavior.
Build and run dictation end-to-end on both a local and cloud-hosted assistant to confirm: (a) successful daemon responses still work within 5s, (b) fallback kicks in promptly on failure and text is inserted correctly.
Check monitoring/alerting: The log level change from error → warning could affect any dashboards or alerts keyed on DictationClient error-level logs.

Notes

This is a companion to platform-side PR #4007 (currently draft) which adds server-side graceful degradation. The two are independent — either can land on its own.
Could not build locally (macOS/Xcode project on Linux VM). Relying on CI for build validation.

Link to Devin session: https://app.devin.ai/sessions/4d99242121f94fed84c1d80e873c569a
Requested by: @tkheyfets

Open with Devin


          [LUM-753] Improve dictation resilience: reduce timeout, use warning-l…

c46f894

…evel logs, add timing

- Reduce dictation HTTP timeout from 10s to 5s so the client falls back
  to raw transcription faster when the daemon is unreachable
- Downgrade failure logs from error to warning since these are expected
  failures with graceful recovery (not actionable errors)
- Add elapsed time to all failure log messages for debugging latency
- Improve fallback log to include transcription length for diagnostics

Co-Authored-By: tkheyfets <timur@vellum.ai>

devin-ai-integration Bot assigned tkheyfets

devin-ai-integration Bot requested a review from tkheyfets

April 7, 2026 18:12

Contributor Author

devin-ai-integration Bot commented Apr 7, 2026

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

Disable automatic comment and CI monitoring

Contributor Author

devin-ai-integration Bot commented Apr 7, 2026

Devin is archived and cannot be woken up. Please unarchive Devin if you want to continue using it.

4 similar comments

Contributor Author

devin-ai-integration Bot commented Apr 7, 2026

Devin is archived and cannot be woken up. Please unarchive Devin if you want to continue using it.

Contributor Author

devin-ai-integration Bot commented Apr 7, 2026

Devin is archived and cannot be woken up. Please unarchive Devin if you want to continue using it.

Contributor Author

devin-ai-integration Bot commented Apr 7, 2026

Devin is archived and cannot be woken up. Please unarchive Devin if you want to continue using it.

Contributor Author

devin-ai-integration Bot commented Apr 7, 2026

Devin is archived and cannot be woken up. Please unarchive Devin if you want to continue using it.

Contributor

tkheyfets commented Apr 7, 2026

@codex review

devin-ai-integration Bot commented

View reviewed changes

Contributor Author

devin-ai-integration Bot left a comment

Devin Review found 2 potential issues.

View 1 additional finding in Devin Review.

Open in Devin Review

clients/shared/Network/DictationClient.swift

clients/shared/Network/DictationClient.swift

+                  /// Timeout for the dictation HTTP request. Kept short so the client falls
+                  /// back to raw transcription quickly when the daemon is unreachable rather
+                  /// than leaving the user staring at a "Processing…" spinner.
+                  static let requestTimeout: TimeInterval = 5

Contributor Author

devin-ai-integration Bot Apr 7, 2026

🚩 Timeout reduction from 10s to 5s may cause more fallback usage under normal latency

The timeout was halved from 10s to 5s. The comment explains the rationale (faster fallback when the assistant is unreachable). However, if the dictation endpoint legitimately takes >5s under load or with longer transcriptions, users will silently get raw transcription instead of the cleaned-up assistant response. The new timing instrumentation will help diagnose this — the elapsed value in warning logs will show whether timeouts cluster near 5s (suggesting the timeout is too tight) vs near 0s (network unreachable). Worth monitoring after deployment.

Open in Devin Review

Was this helpful? React with 👍 or 👎 to provide feedback.

chatgpt-codex-connector Bot reviewed

View reviewed changes

chatgpt-codex-connector Bot left a comment

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: c46f894c45

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

clients/shared/Network/DictationClient.swift Outdated

                   ]
+                  /// Timeout for the dictation HTTP request. Kept short so the client falls
+                  /// back to raw transcription quickly when the daemon is unreachable rather

chatgpt-codex-connector Bot Apr 7, 2026

Replace daemon terminology in client code comments

This comment reintroduces the term daemon in client code, which violates the explicit rule in clients/AGENTS.md (“Terminology: Avoid "Daemon" in Client Code”). The clients guidance requires using “assistant” in client-facing naming/comments to keep terminology consistent and avoid leaking backend implementation detail into client surfaces; please update this wording accordingly.

Useful? React with 👍 / 👎.

tkheyfets requested review from dvargasfuertes and noanflaherty

April 7, 2026 18:19

tkheyfets previously approved these changes

View reviewed changes


          Apply suggestions from code review

f5c41a0

Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com>

noanflaherty dismissed tkheyfets’s stale review via

f5c41a0

April 7, 2026 18:21

noanflaherty merged commit b80255c into main

3 checks passed

noanflaherty deleted the devin/1775585427-LUM-753-dictation-resilience branch

April 7, 2026 18:21

devin-ai-integration Bot mentioned this pull request

[LUM-753] Add diagnostic logging to dictation recording lifecycle #24067

Merged

3 tasks

noanflaherty added a commit that referenced this pull request


          [LUM-753] Improve client-side dictation resilience (#24042)

4a2c641

* [LUM-753] Improve dictation resilience: reduce timeout, use warning-level logs, add timing

- Reduce dictation HTTP timeout from 10s to 5s so the client falls back
  to raw transcription faster when the daemon is unreachable
- Downgrade failure logs from error to warning since these are expected
  failures with graceful recovery (not actionable errors)
- Add elapsed time to all failure log messages for debugging latency
- Improve fallback log to include transcription length for diagnostics

Co-Authored-By: tkheyfets <timur@vellum.ai>

* Apply suggestions from code review

Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com>

---------

Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Co-authored-by: tkheyfets <timur@vellum.ai>
Co-authored-by: Noa Flaherty <noa@vellum.ai>

noanflaherty mentioned this pull request

Cherry-pick fixes into release/v0.6.2 #24074

Merged

2 tasks

noanflaherty added a commit that referenced this pull request


          Cherry-pick fixes into release/v0.6.2 (#24074)

c083899

* fix: use borderElement for unchecked checkbox visibility (LUM-751) (#23992)

The ToS checkbox border used VColor.borderBase (#24292E dark) which is
identical to VCard's VColor.surfaceLift background (#24292E dark),
making the unchecked checkbox completely invisible in dark mode.

Switch to VColor.borderElement (#5A6672 dark / #CFCCC9 light) which
is the semantic token for interactive element borders and provides
clear contrast against the card background in both color schemes.

Closes LUM-751

Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Co-authored-by: ashlee@vellum.ai <ashlee@vellum.ai>

* fix: enable link click handling in VSelectableTextView (#23986)

Add NSTextViewDelegate to Coordinator with textView(_:clickedOnLink:at:)
to open clicked links in the default browser via NSWorkspace. The NSTextView
was already styling links (tint color, underline, pointer cursor) but had
no delegate to handle click events, making links visually styled but inert.

Closes LUM-748

Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Co-authored-by: ashlee@vellum.ai <ashlee@vellum.ai>

* fix: prevent onboarding from repeating name confirmation after tool calls (#23989)

The agentic loop calls the LLM again after tool results return. Combined
with "talk before you work" (SOUL.md) and "save immediately" (BOOTSTRAP.md),
the model would confirm the user's name, call file_edit, then re-confirm
the name in the continuation response.

Add explicit guidance in both BOOTSTRAP.md and SOUL.md to not repeat text
that was already shown before tool calls.

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* Update Vellum Cloud onboarding copy to emphasize always-on availability (#24022)

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* [LUM-753] Improve client-side dictation resilience (#24042)

* [LUM-753] Improve dictation resilience: reduce timeout, use warning-level logs, add timing

- Reduce dictation HTTP timeout from 10s to 5s so the client falls back
  to raw transcription faster when the daemon is unreachable
- Downgrade failure logs from error to warning since these are expected
  failures with graceful recovery (not actionable errors)
- Add elapsed time to all failure log messages for debugging latency
- Improve fallback log to include transcription length for diagnostics

Co-Authored-By: tkheyfets <timur@vellum.ai>

* Apply suggestions from code review

Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com>

---------

Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Co-authored-by: tkheyfets <timur@vellum.ai>
Co-authored-by: Noa Flaherty <noa@vellum.ai>

---------

Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Co-authored-by: ashlee@vellum.ai <ashlee@vellum.ai>
Co-authored-by: asharma53 <64060709+asharma53@users.noreply.github.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-authored-by: tkheyfets <timur@vellum.ai>

vellum-automation Bot added a commit that referenced this pull request


          Release v0.6.2 (#24108)

83c24d6

* Release v0.6.2

* Cherry-pick fixes into release/v0.6.2 (#24074)

* fix: use borderElement for unchecked checkbox visibility (LUM-751) (#23992)

The ToS checkbox border used VColor.borderBase (#24292E dark) which is
identical to VCard's VColor.surfaceLift background (#24292E dark),
making the unchecked checkbox completely invisible in dark mode.

Switch to VColor.borderElement (#5A6672 dark / #CFCCC9 light) which
is the semantic token for interactive element borders and provides
clear contrast against the card background in both color schemes.

Closes LUM-751

Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Co-authored-by: ashlee@vellum.ai <ashlee@vellum.ai>

* fix: enable link click handling in VSelectableTextView (#23986)

Add NSTextViewDelegate to Coordinator with textView(_:clickedOnLink:at:)
to open clicked links in the default browser via NSWorkspace. The NSTextView
was already styling links (tint color, underline, pointer cursor) but had
no delegate to handle click events, making links visually styled but inert.

Closes LUM-748

Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Co-authored-by: ashlee@vellum.ai <ashlee@vellum.ai>

* fix: prevent onboarding from repeating name confirmation after tool calls (#23989)

The agentic loop calls the LLM again after tool results return. Combined
with "talk before you work" (SOUL.md) and "save immediately" (BOOTSTRAP.md),
the model would confirm the user's name, call file_edit, then re-confirm
the name in the continuation response.

Add explicit guidance in both BOOTSTRAP.md and SOUL.md to not repeat text
that was already shown before tool calls.

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* Update Vellum Cloud onboarding copy to emphasize always-on availability (#24022)

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* [LUM-753] Improve client-side dictation resilience (#24042)

* [LUM-753] Improve dictation resilience: reduce timeout, use warning-level logs, add timing

- Reduce dictation HTTP timeout from 10s to 5s so the client falls back
  to raw transcription faster when the daemon is unreachable
- Downgrade failure logs from error to warning since these are expected
  failures with graceful recovery (not actionable errors)
- Add elapsed time to all failure log messages for debugging latency
- Improve fallback log to include transcription length for diagnostics

Co-Authored-By: tkheyfets <timur@vellum.ai>

* Apply suggestions from code review

Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com>

---------

Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Co-authored-by: tkheyfets <timur@vellum.ai>
Co-authored-by: Noa Flaherty <noa@vellum.ai>

---------

Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Co-authored-by: ashlee@vellum.ai <ashlee@vellum.ai>
Co-authored-by: asharma53 <64060709+asharma53@users.noreply.github.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-authored-by: tkheyfets <timur@vellum.ai>

---------

Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: Noa Flaherty <noa@vellum.ai>
Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Co-authored-by: ashlee@vellum.ai <ashlee@vellum.ai>
Co-authored-by: asharma53 <64060709+asharma53@users.noreply.github.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-authored-by: tkheyfets <timur@vellum.ai>

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

chatgpt-codex-connector[bot] chatgpt-codex-connector[bot] left review comments

dvargasfuertes Awaiting requested review from dvargasfuertes

noanflaherty Awaiting requested review from noanflaherty

+1 more reviewer

tkheyfets tkheyfets left review comments

Labels

None yet