AI Suggestion Core

[Omoshirokunai]

What

Implement the non-blocking suggestion generation system in the service worker.

Core Components

1. Suggestion Queue (src/lib/ai/suggestion-queue.js)

• Debounces keystroke events at 500ms (user perceives as instant)
• Routes suggestions through fast brain (instant) then slow brain (async)
• Stores results in chrome.storage.local.suggestions
• Fires SUGGESTIONS_READY message to UI

2. Ollama Client (src/lib/ai/ollama-client.js)

• Async HTTP client to http://localhost:11434/api/generate
• 2000ms timeout and then fast fallback to second brain
• Provides: sentence completion, paraphrase
• Handles connection failures silently
• No blocking on the calling thread

3. second brain (src/lib/ai/heuristic-brain.js)

• Returns suggestions in <5ms
• Provides:
  • Word completion predictions (no AI)
  • Synonym detection (word frequency lookup)
  • Grammar error detection (regex)
  • Pattern detection (e.g., "According to...")
• Gracefully handles empty/short input

Acceptance Criteria

• SuggestionQueue queues with 500ms debounce
• HeuristicBrain.suggest() returns in <5ms, no async calls
• OllamaClient.suggest() has 2000ms timeout
• Suggestions stored in chrome.storage.local.suggestions
• Service worker integration: LOG_EVENT 'edit' triggers enqueue (no await)
• GET_SUGGESTIONS message handler returns stored suggestions
• No typing lag observed (content script unaffected)