AI Chat & Code Generation

Chat Interface

The chat panel is your main tool for interacting with the AI agent:

• Type requests in natural language (any language)
• See streaming responses in real-time
• View tool calls (Read, Write, Edit, Search, etc.) as they happen
• Track cost per interaction in the sidebar

What the AI Can Do

• Generate code — Create entire pages, components, and features from a description
• Edit existing code — Modify specific files, functions, or styles
• Search and analyze — Find patterns across your project, analyze images
• Fix bugs — Describe the problem, the AI reads error logs and fixes it
• Generate images — Create custom AI-generated images for your project
• Manage backend — Create API endpoints, database tables, and migrations

Choosing the AI Model

The model picker in the chat decides which AI model runs your requests:

• Claude (Anthropic) — the default
• GPT (OpenAI)
• Gemini (Google)
• DeepSeek
• Mistral

The choice is saved per project and applies to every chat in it. Each entry shows relative cost, speed and quality so you can match the model to the task. All models are billed in the same AI Tasks.

Plan Confirmation

For new projects, the AI first generates a plan showing what it will build. Review and click Yes, start to proceed, or No, refine to adjust before generation begins.

File Attachments

Upload files directly in the chat:

• Images — Used as design reference or added to the project
• PDFs — The AI can read and extract content from documents
• Other files — DOCX, XLSX, PPTX are supported

Drag and drop files into the chat area or use the upload button.

Multiple Chats

Create separate chats for different features. Each chat has its own:

• Conversation history
• Git branch (when GitHub is connected)
• Context and state

Context Window

The AI maintains context of your entire project — it knows all files, the current state, and previous conversations. For very long conversations, XAEIO automatically compacts older messages to keep the context efficient.

Stopping Generation

Click the Stop button to interrupt the AI at any time. Partial work is preserved — the files already written stay in your project.

Following what the agent does

While the agent works, its tool calls appear in the chat grouped by operation rather than as a flat list of steps, each with a name that says what it did. File generation shows up the same way, inside the agent's block, so you can see which files a turn produced without opening the file browser.

If a review turns up problems, they are listed in the chat with their details instead of only appearing as a failed step.