AI Chat

ChatML's AI chat provides real-time interaction with Claude, giving it direct access to your codebase.

Real-Time Streaming

When Claude works, you see everything as it happens:

• Text streaming — Response text appears as it's generated, rendered as Markdown with syntax highlighting
• Tool execution — Each tool call shows its name, parameters, status (running/success/failure), and duration
• Extended thinking — When enabled, Claude's reasoning process is displayed in a collapsible section
• Sub-agent tracking — When Claude spawns sub-agents for parallel tasks, each is tracked independently
• Elapsed time — A running timer shows how long the current response has been generating

Sending Messages

Type your message in the input area and press Enter to send. Claude receives your message along with the full conversation context and begins working.

Message Features

• Markdown formatting — Messages support full Markdown
• File attachments — Attach code files or images for context
• Multi-line input — Use Shift+Enter for new lines
• Message recall — Press Up Arrow in an empty input to recall your last message

Extended Thinking

Extended thinking allows Claude to reason through complex problems before responding. When enabled:

1. Claude's internal reasoning appears in a collapsible "Thinking" section
2. The final response is more thorough and well-considered
3. Configurable token budget controls how much thinking is allowed

Enable extended thinking in Settings > AI > Extended Thinking.

Stopping the Agent

• Cmd+Shift+S — Stop the current agent
• Cmd+. — Interrupt the agent (softer stop)
• Click the Stop button in the UI

Model Selection

Choose your Claude model in Settings > AI > Model:

• Claude Opus — Most capable, best for complex tasks
• Claude Sonnet — Balanced capability and speed
• Claude Haiku — Fastest, best for simple tasks

You can also set a fallback model that activates when the primary model fails.