Chatting with the AI Research Agent

When I have questions about AI infrastructure, I want to chat with an AI that knows my sources, so I can get recommendations grounded in authoritative documentation.

---

Introduction

The Lab panel is where research happens. Unlike generic AI assistants, Lattice’s research agent is grounded in your specific sources. Every recommendation includes numbered citations linking back to specific chunks in your documents.

How Grounded Chat Works

1. You ask a question — Type naturally, as you would with any AI assistant
2. Lattice searches your sources — Semantic and keyword search finds relevant chunks
3. The agent reasons — LangGraph-powered multi-step reasoning synthesizes findings
4. Citations included — Every claim links to a specific source chunk

Using @ Mentions

Reference specific sources in your questions using @ mentions:

“Compare the pricing in @anthropic-pricing with @openai-pricing for a 1M token/day workload”

This focuses the agent on specific documents for more targeted analysis.

Transparent Reasoning

Watch the agent’s thinking process:

• Thinking steps show which sources are being searched
• Reasoning traces explain how conclusions are reached
• Citation chains link claims to evidence

What You Can Ask

• Model comparisons: “Compare Claude Sonnet vs GPT-4 for RAG workloads”
• Pricing analysis: “What would 100K requests/day cost on Bedrock vs Vertex?”
• Architecture recommendations: “What’s the best embedding model for legal documents?”
• Compliance questions: “Which providers offer HIPAA-compliant inference?”

Every response includes numbered citations [1], [2], [3] that you can click to verify.

Next steps: Save valuable responses as artifacts in the Studio panel.