AI Chat Interface

The AI Chat is a conversational interface that lets you search for experts, check conflicts, pull up profiles, and manage cases using natural language... no forms or menus required.

Getting started

Open Chat from the sidebar. You'll see a message input at the bottom and a conversation history sidebar on the left.

Type your question or request as you would to a knowledgeable colleague. For example:

I need a neurologist in Irvine who works mostly for the defense and has testified before.

The AI will interpret your request, call the appropriate tools, and return structured results with explanations. If no conversation exists yet, one is created automatically when you send your first message.

What the AI can do

The assistant has access to specialized tools for expert matching:

Model selection

The chat input includes a model selector in the toolbar. Models are grouped by provider:

Anthropic

OpenAI

Perplexity (Research)

The model resets to the default when you switch conversations.

Extended Thinking

Some models support Extended Thinking, which lets the AI show its step-by-step reasoning before providing a final answer. Supported models:

• Claude Opus 4.6

• GPT-5.2

When you select a supported model, a thinking toggle appears at the bottom of the model dropdown. Thinking is auto-enabled for models that support it and auto-disabled otherwise.

During streaming, the AI's thinking process appears in a collapsible block above the response. You can expand or collapse it at any time.

Example prompts

Find me an orthopedic surgeon near El Segundo, California

I need a cardiologist within 20 miles of Irvine who has testified at least once, works 70%+ for plaintiff side, and is a LITILI exclusive

Check if Dr. Smith (expert ID: EXP-1234) has any conflicts with plaintiff Johnson & Johnson or defendant law firm Baker McKenzie

Show me the full profile for expert EXP-5678

Find experts similar to Dr. Martinez but closer to Houston

What are the latest trends in medical expert witness testimony for cardiology malpractice cases?

Switch to a Research model (Sonar Deep Research, Sonar Pro, or Sonar) for questions outside of the Litili database.

Tool call visibility

When the AI uses a tool, you'll see:

1. Tool call start — a card showing which tool is being called and the input parameters, with a "running" status indicator

2. Tool call update — the status changes to "completed" (or "error") and shows the output data (expert lists, profiles, conflict status)

3. AI interpretation — the assistant's summary and recommendations based on the tool results

While the AI is deciding which tools to use (before any tool call or content appears), a skeleton loader is shown to indicate activity.

This transparency lets you verify the AI's reasoning and the underlying data.

Conversations

Sidebar

The left sidebar shows your conversation history, sorted by most recent activity. Each entry displays:

• The conversation title (auto-generated or "Untitled chat")

• A relative timestamp ("Updated 2 hours ago")

Toggle the sidebar with the sidebar button in the header. Its visibility persists across sessions.

Managing conversations

• New chat — Click the + button in the sidebar header, or the conversation resets when you navigate away from an active conversation

• Switch — Click any conversation in the sidebar to load it

• Delete — Hover over a conversation and click the trash icon that appears

Context continuity

Context carries across messages within a conversation. You can say "now check that expert for conflicts" after a search without re-specifying who you mean. Conversations are stored server-side and persist across sessions.

Stopping a response

While a response is streaming, the send button transforms into a stop button (red circle). Click it to cancel generation immediately.

Guardrails

The AI operates within defined safety boundaries:

If a guardrail is hit, you'll see a clear error message. Simply rephrase or simplify your request.

Error handling

The chat provides contextual error messages based on the type of failure:

Transient errors (5xx, timeouts) are automatically retried by the backend gateway. You'll see a warning toast showing the retry attempt count.