Lab (Chat)

The Lab is Lattice’s AI-powered chat interface where you interact with an intelligent assistant that grounds every response in your curated sources.

How It Works

When you send a message, Lattice:

1. Searches your sources using hybrid retrieval
2. Analyzes the query to identify relevant context
3. Reasons through the information with transparent thinking steps
4. Responds with cited, verifiable claims

Grounded Responses

Every response from Lattice includes numbered citations:

Based on the documentation [1], Claude Sonnet offers a 200K
context window [2], while GPT-4 Turbo supports 128K tokens [3].
For high-volume applications, the pricing differs significantly [4].

[1] Anthropic Documentation - Model Overview
[2] Anthropic Documentation - Context Windows
[3] OpenAI Documentation - GPT-4 Turbo
[4] Anthropic Pricing - API Rates

Click any citation to see the original source passage.

Thinking Steps

Lattice exposes its reasoning process through thinking steps:

[Thinking] Analyzing the query about context windows...
[Thinking] Found 3 relevant sources with specifications...
[Thinking] Comparing Claude and GPT-4 context limits...
[Thinking] Checking pricing implications for large contexts...

Tip

Thinking steps help you verify the AI’s reasoning and catch any gaps in logic or source coverage.

Suggested Prompts

The Lab provides context-aware suggested prompts based on:

• Your current workspace sources
• Recent conversation topics
• Common research patterns

Click any suggestion to use it as your next query.

Chat Modes

Research Agent (Default)

The full-featured mode with:

• Transparent thinking steps
• Multi-source synthesis
• Citation tracking
• Artifact detection

Direct LLM

Bypass the research agent for simple queries:

• Faster responses
• No source grounding
• Useful for general questions

Using Context Effectively

@Mentions for Source Boosting

Reference specific sources to prioritize them:

@anthropic-docs What are the rate limits for Claude?

Scenario Context

Set an active scenario to inform responses:

Given my high-volume chat scenario, which model
offers the best latency/cost tradeoff?

Stack Context

Reference stack configurations for targeted advice:

Using my Claude Haiku Speed Stack, how should I
configure the temperature for consistency?

Conversation History

The Lab maintains full conversation history within each workspace. You can:

• Scroll back to review previous exchanges
• Search history for specific topics
• Delete individual messages
• Export conversations for documentation

Note

Conversation history is used to maintain context across messages. Configure the history limit in Settings to control how much context is included.

Streaming Responses

Responses stream in real-time via Server-Sent Events (SSE):

event: step
data: {"type": "thinking", "content": "Analyzing query..."}

event: content
data: {"type": "text", "content": "Based on the documentation"}

event: content
data: {"type": "text", "content": " [1], Claude Sonnet..."}

event: done
data: {"type": "complete", "usage": {"input_tokens": 1234, "output_tokens": 567}}

API Reference

Send Message

Streaming

POST /api/workspaces/{workspace_id}/chat
Content-Type: application/json

{
  "message": "Compare Claude and GPT-4 for RAG applications",
  "stream": true,
  "include_history": true,
  "scenario_id": "optional-scenario-uuid",
  "stack_id": "optional-stack-uuid"
}

Returns: Server-Sent Events stream

Non-Streaming

POST /api/workspaces/{workspace_id}/chat
Content-Type: application/json

{
  "message": "Compare Claude and GPT-4 for RAG applications",
  "stream": false
}

Returns: JSON response with complete message

List Messages

GET /api/workspaces/{workspace_id}/messages

Delete Message

DELETE /api/workspaces/{workspace_id}/messages/{message_id}