Skip to main content

AI Chat

Cync provides AI-powered chat interfaces for querying data using natural language.

Interfaces​

Rhea Chat (/protected/rhea)​

General-purpose AI assistant with access to:

  • Stickies
  • Links
  • News articles

Fireflies Chat​

Embedded in transcript viewer for meeting-specific queries.

How It Works​

┌────────────┐    ┌────────────┐    ┌────────────┐
│   User     │───▶│  OpenRouter│───▶│    MCP     │
│   Query    │    │    LLM     │    │   Tools    │
└────────────┘    └────────────┘    └──────┬─────┘
                                          │
                        ┌─────────────────┼─────────────────┐
                        ▼                 ▼                 ▼
                   ┌─────────┐      ┌─────────┐      ┌─────────┐
                   │ Stickies│      │  Links  │      │  News   │
                   └─────────┘      └─────────┘      └─────────┘

MCP Tools​

The chat uses Model Context Protocol (MCP) tools:

ToolDescription
get_stickiesRetrieve user's notes
create_stickyCreate a new note
edit_stickyUpdate existing note
search_newsSearch news articles
get_weekly_news_digestSummarize recent news

Example Queries​

Stickies​

"Show me my notes from last week"
"Create a sticky about the Q4 planning meeting"
"Find notes tagged with 'urgent'"

News​

"What's happening in the tech industry?"
"Summarize the news from this week"
"Find articles about AI regulation"

Meetings (Fireflies Chat)​

"What were the action items from this meeting?"
"Who talked the most?"
"What did we decide about the budget?"

Multi-Model Support​

OpenRouter provides access to multiple AI models:

# Choose your model in .env.local
OPENROUTER_MODEL=anthropic/claude-3.5-sonnet
# OR
OPENROUTER_MODEL=openai/gpt-4o
# OR
OPENROUTER_MODEL=meta-llama/llama-3.1-70b-instruct

Feedback System​

Chat interfaces include thumbs up/down for feedback:

  • Currently logs to console
  • Future: Store feedback for model improvement

Architecture​

Chat API Route​

// /api/chat/route.ts
export async function POST(request: Request) {
  const { messages } = await request.json();

  // Get user context from session
  const user = await getUser();

  // Call OpenRouter with tools
  const response = await openrouter.chat({
    model: process.env.OPENROUTER_MODEL,
    messages,
    tools: getMcpTools(user.id)
  });

  return Response.json(response);
}

Tool Execution​

// When LLM decides to use a tool
if (response.tool_calls) {
  for (const call of response.tool_calls) {
    const result = await executeTool(call.name, call.arguments);
    // Feed result back to LLM
  }
}