Skip to main content

What are AI Agents?

AI agents are specialized AI assistants that generate and modify code based on your natural language instructions. Each agent has unique capabilities optimized for specific tasks.

Agent Types

Tesslate Studio has two core agent types:

Stream Agent

Best for: UI building, fast iteration
  • Real-time streaming responses
  • Generates code continuously
  • Perfect for visual components
  • Faster for simple tasks

Iterative Agent

Best for: Complex logic, multi-step tasks
  • Step-by-step reasoning with tool calling
  • Reads, writes, and edits files
  • Executes shell commands
  • Better for complex features

Agent Tools

Agents have access to powerful tools for interacting with your project:
ToolDescription
read_fileRead contents of any file in the project
write_fileCreate or overwrite files
edit_fileMake targeted edits to specific sections
list_filesBrowse project directory structure
Tool usage requires approval. The agent will ask for confirmation before executing commands that modify files or run shell commands.

Agent Characteristics

Every agent has:
  • Icon & Name: Visual identification
  • System Prompt: Instructions that guide behavior
  • Model: The AI model powering the agent (GPT-4, Claude, etc.)
  • Mode: Stream or Iterative processing
  • Source Type: Open source or closed source

Agent Lifecycle

1

Discover

Browse agents in the Marketplace
2

Install

Add agents to your Library
3

Enable

Toggle agents on/off in your library
4

Use

Select agents in project chat to generate code
5

Customize

Edit prompts and models for custom agents

Using Agents in Projects

Selecting an Agent

In any project, you can choose which agent to use:
  1. Open the chat interface
  2. Click the agent selector dropdown
  3. Choose from your enabled agents
  4. Start chatting!

When to Use Each Type

Use Stream Agent when:
  • Building UI components
  • Creating landing pages
  • Making quick visual changes
  • Iterating rapidly on design
  • Need fast responses

Agent Marketplace

The Marketplace offers:
  • Official Agents: Built and maintained by Tesslate
  • Community Agents: Created and shared by users
  • Specialized Agents: Focused on specific frameworks or tasks

Categories

  • Builder: General-purpose building agents
  • Frontend: UI-focused agents
  • Full-stack: End-to-end development
  • Data & AI: Data processing and ML

Custom Agents

You can create custom agents by:
  1. Forking an existing open-source agent
  2. Creating from scratch in your library
  3. Customizing the system prompt
  4. Selecting the AI model
  5. Publishing to the community (optional)
Custom agents let you specialize behavior for your specific needs, like using particular libraries or coding styles.

Agent Models

Agents can use different AI models:
  • GPT-4: OpenAI’s most capable model
  • Claude: Anthropic’s advanced model
  • Qwen: Fast, efficient coding model
  • Custom Models: Via OpenRouter integration
For closed-source agents, the model is fixed. For open-source agents, you can change the model in your library.

Best Practices

  • Use Stream for UI, Iterative for logic
  • Switch agents mid-project as needed
  • Try different agents to find what works
  • Describe exactly what you want
  • Include technology preferences
  • Mention styling requirements
  • Specify behavior details
  • Build features one at a time
  • Test after each change
  • Refine through conversation
  • Don’t try to do everything at once

Next Steps