Skip to main content
The Whilst AI Assistant (Sidekick) is a chat interface built into the web app that provides intelligent document management through natural language.

​
Capabilities

  • Document Search & Discovery — Semantic search with citation of relevant sources
  • Document Operations — Create, edit, organize, summarize, and analyze docs via chat
  • Conversation Management — Persistent history, automatic titling, context awareness
  • Surgical Editing — Targeted changes to specific document sections without rewriting

​
Chat Pipeline

The assistant uses a parallelized pipeline to minimize time-to-first-token: 
User Message
    ↓
1. Create/Load Conversation
    ↓
2. [PARALLEL]
   ├─ Enrich Context
   ├─ Get Recent Messages
   └─ Smart Document Search
    ↓
3. [ASYNC/NON-BLOCKING]
   ├─ Classify Intent
   └─ Generate Title
    ↓
4. Build AI Context → Stream Response
Performance: ~1s time-to-first-token, ~2.5s total response time. The assistant skips unnecessary searches to save latency:
  • Always searches on first message in a conversation
  • Skips if the query is too short (likely a follow-up)
  • Skips if the user is viewing a specific document (uses that as context)

​
Surgical Edit System

Applies targeted changes to documents without rewriting the entire content: 
const result = await applySurgicalEdit(
  currentContent,
  "Reformat the ingredients into a single list",
  docTitle
);
// Only the targeted section is modified
Rules enforced by the system prompt:
  1. Only modify the specific part mentioned
  2. Preserve 100% of all other content
  3. Return the complete document with targeted changes

​
Frontend

The chat UI is located at apps/web-app/components/docs-ai-chat-enhanced.tsx and features:
  • Minimal single-panel design inspired by Cursor
  • Chat history dropdown for resuming conversations
  • Real-time streaming via Server-Sent Events
  • Inline citations with similarity scores
  • Custom loading animation (Whilst logo “running” effect)
  • Keyboard shortcuts: Cmd+U for files, Cmd+Shift+V for voice