Overview
The agent uses tools to perform actions. Tools are either non-terminal (results feed back to the agent) or terminal (execution completes).
| Category | Tools | Type |
|---|
| Search | search_models, get_model_details, web_search | Non-terminal |
| Analysis | vision_preprocessor | Non-terminal |
| Execution | execute_model, generate_text | Terminal |
| Workflow | create_workflow, trigger_workflow, build_workflow, check_execution | Mixed |
| Interaction | ask_clarification | Terminal |
search_models
Find AI models matching a use case.
// Input
{ "use_case": "generate portrait photo from text" }
// Output
{
"content": "Found 5 models for portrait generation...",
"metadata": {
"tool": "search_models",
"count": 5,
"models": [
{ "name": "nano-banana-pro", "category": "text-to-image", "score": 0.95 },
{ "name": "flux-2-max", "category": "text-to-image", "score": 0.92 }
]
}
}
get_model_details
Get input schema for a specific model.
// Input
{ "model_name": "nano-banana-pro" }
// Output
{
"content": "nano-banana-pro - Text to Image Generation",
"metadata": {
"tool": "get_model_details",
"input_schema": {
"prompt": { "type": "string", "required": true },
"aspect_ratio": { "type": "string", "enum": ["1:1", "16:9", "9:16"] },
"negative_prompt": { "type": "string" }
}
}
}
web_search
Search the web for information.
// Input
{ "query": "latest AI image generation techniques", "recency": "month" }
execute_model
Run an AI model with specified parameters. Terminal, so it returns the result directly.
// Input
{
"model_name": "nano-banana-pro",
"inputs": {
"prompt": "Professional headshot portrait",
"aspect_ratio": "1:1"
}
}
// Output
{
"content": "Image generated successfully",
"metadata": {
"tool": "execute_model",
"model": "nano-banana-pro",
"generations": ["https://storage.eachlabs.ai/portrait.png"],
"execution_time_ms": 12500
}
}
vision_preprocessor
Analyze image content before processing.
// Input
{ "image_urls": ["https://example.com/photo.jpg"] }
// Output
{
"content": "Image analysis: Portrait photo, female, natural lighting...",
"metadata": { "tool": "vision_preprocessor", "detected_objects": [...] }
}
create_workflow
Create a new multi-step workflow.
trigger_workflow
Start execution of a workflow. Non-terminal, so the agent keeps monitoring execution.
build_workflow
Build or update a workflow definition. Terminal.
check_execution
Check status and results of a workflow execution. Terminal.
ask_clarification
Ask the user for more information. Terminal, yielding a clarification_needed event and waiting for a response.
// Input
{
"question": "What style would you like?",
"options": ["Photorealistic", "Artistic", "Anime", "Cinematic"],
"context": "Different styles produce very different results."
}
Image Generation
1. search_models("portrait generation") → finds nano-banana-pro
2. get_model_details("nano-banana-pro") → gets input schema
3. execute_model("nano-banana-pro", {...}) → generates image [TERMINAL]
Image Analysis + Edit
1. vision_preprocessor([image_url]) → analyzes image
2. search_models("remove background") → finds model
3. execute_model("eachlabs-bg-remover-v1") → removes bg [TERMINAL]
Workflow Creation + Execution
1. search_models("portrait generation") → finds models
2. search_models("image to video") → finds more models
3. create_workflow({...}) → creates workflow
4. trigger_workflow(...) → starts execution
5. check_execution(...) → monitors progress [TERMINAL]
Last modified on March 6, 2026
