OpenRouter provider, latency fixes, and permission hardening#22
Open
christmasday wants to merge 3 commits into
Open
OpenRouter provider, latency fixes, and permission hardening#22christmasday wants to merge 3 commits into
christmasday wants to merge 3 commits into
Conversation
Adds OpenRouter as a second LLM provider alongside Gemini, with full
routing in the LLM service, onboarding wizard, and settings UI.
Key changes:
- New _openrouter* methods in LLMService for image/text/transcription
processing, using raw HTTPS (OpenAI-compatible API)
- Provider routing guards in all public processing methods
- Onboarding wizard: LLM provider choice step with conditional
Gemini/OpenRouter key fields, summary display
- Settings UI: provider dropdown with conditional field groups
- first-run.js: getStatus() returns llmProvider, openrouterConfigured,
openrouterModel; needsOnboarding() checks both providers
- macOS permission descriptions (NSMicrophoneUsageDescription,
NSScreenCaptureDescription, NSCameraUsageDescription)
Performance improvements:
- Parallelized dual-method retry (_raceGeminiMethods) — race SDK vs HTTPS
instead of sequential fallback
- Reduced config defaults: timeout 60000→30000ms, maxRetries 3→1
- Increased https maxSockets 1→4
- Removed 500ms artificial delays in main.js (lines 366, 583 -> 50ms)
- Removed 200ms window rendering delays in llm-response-window.js
- Removed 100ms display verification setTimeout
- Removed 200ms resize request setTimeout
Stability fixes:
- try-catch around desktopCapturer.getSources() to prevent macOS 14+
crash when screen recording permission is missing
- systemPreferences.getMediaAccessStatus('screen') pre-check with clear
error message
- _ensureMicrophonePermission() using askForMediaAccess('microphone')
guards all recording entry points
- isDestroyed() guard in showLLMLoading() to prevent sending IPC to a
destroyed window
|
@christmasday is attempting to deploy a commit to the csrsoftwares' projects Team on Vercel. A member of the Team first needs to authorize it. |
Promise.allSettled waited for ALL promises to settle before returning the fastest result. This caused the user to wait for the slowest method (e.g. 30s HTTPS timeout) even when the SDK method succeeded in 2s. Promise.any returns as soon as the FIRST method fulfills, so there is zero latency penalty from the slower method on the happy path.
BREAKING: The previous single-skill (DSA) system is replaced by three distinct modes that users can navigate via keyboard shortcuts or settings. New skills: - General: versatile assistant for a wide variety of questions - Coding: programming-focused mode with language enforcement - Meeting: passive listener that detects questions during meetings Changes: - prompts/: replaced dsa.md + unused programming.md with general.md, coding.md, meeting.md - prompt-loader.js: - skillsRequiringProgrammingLanguage: dsa -> coding - Loads all three .md prompts (drops the dsa-only filter) - normalizeSkillName maps old names to new skills - getAvailableSkills returns [general, coding, meeting] - main.js: default skill 'general'; navigateSkill cycles 3 skills; programming language gates on 'coding' - session.manager.js: default skill 'general' - src/ui/main-window.js: skill names, icons, language selector visibility (shown only in Coding mode), click handler cycles forward - settings.html: dropdown options for three skills - src/ui/chat-window.js: updated skill icon map - src/services/llm.service.js: formatImageInstruction, formatUserMessage, getIntelligentTranscriptionPrompt updated per mode; fallback responses and skill keywords updated for new skills
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds OpenRouter as a second LLM provider alongside Gemini, with full provider routing throughout the app. Onboarding and settings UI now include provider choice. Also includes performance improvements and stability hardening.
OpenRouter Provider
_openrouter*methods inLLMServicefor image/text/transcription processing via raw HTTPS (OpenAI-compatible API)processImageWithSkill,processTextWithSkill,processTranscriptionWithIntelligentResponse,testConnection,getStats,updateApiKeyopenrouter/free(auto-routes to free vision-capable models)first-run.js:getStatus()returnsllmProvider,openrouterConfigured,openrouterModel;needsOnboarding()checks both providersPerformance
_raceGeminiMethods): SDK + HTTPS methods race simultaneously instead of sequential fallback — first success wins, zero latency penalty on happy pathtimeout60000→30000ms,maxRetries3→1main.js(lines 366, 583)llm-response-window.jssetTimeoutin display state verificationsetTimeoutbefore window resize requestStability / Crashes
try/catcharounddesktopCapturer.getSources()+systemPreferences.getMediaAccessStatus(screen)pre-check with clear error message if permission is denied_ensureMicrophonePermission()usingsystemPreferences.askForMediaAccess(microphone)guards all three recording entry points (IPC handle, IPC listener, Alt+R shortcut)isDestroyed()guard inshowLLMLoading()prevents sending IPC to a destroyed windowNSMicrophoneUsageDescription,NSScreenCaptureDescription,NSCameraUsageDescription) inbuild.extendInfo