Agents
Agents are the core execution units in Synatra. Each agent is an AI executor with versioned configurations.
Overview
Section titled “Overview”An agent consists of:
- Working copy - Draft state for iterating on configurations
- Releases - Published, immutable versions with semantic versioning
- Model configuration - Provider, temperature, system prompt
- Assigned tools - Functions the agent can call
Creating an Agent
Section titled “Creating an Agent”- Navigate to Agents in the Console
- Click New Agent
- Configure the model settings
- Add tools as needed
- Write a system prompt
Model Configuration
Section titled “Model Configuration”Synatra supports multiple LLM providers:
| Provider | Models | Special Features |
|---|---|---|
| OpenAI | GPT-4o, GPT-4 Turbo | Reasoning effort control |
| Anthropic | Claude 3.5 Sonnet, Claude 4 | Extended thinking |
| Gemini 1.5 Pro, Gemini 2 | Thinking config |
Releases
Section titled “Releases”When you’re ready to deploy an agent:
- Review the working copy configuration
- Click Create Release
- The release becomes the active version
Releases are immutable and use semantic versioning (major.minor.patch). To make changes, update the working copy and create a new release.
UI Widgets
Section titled “UI Widgets”Agents can generate dynamic UI widgets to interact with users:
| Widget Type | Description |
|---|---|
| Form | Input forms with JSON Schema validation |
| Table | Sortable tables with row selection |
| Chart | Line, bar, and pie charts |
| Markdown | Formatted text content |
| Confirmation | Approve/reject dialogs |
| Key-Value | Display structured data |
Widgets can wait for user input before the agent continues, allowing for interactive approval flows and data collection.