Models
The Models page in the PinionAI Studio interface is dedicated to defining and managing the Large Language Models (LLMs) available to your agents. This allows for fine-grained control over model versions, providers, local hosting configurations, cost tracking, and rate limits.
Prerequisites
Before you can manage Models, ensure the following conditions are met:
- You must be logged into the application.
- Your user role must have the necessary permissions (
owner,admin,editor, ormodels). - You must have an Account selected from the sidebar.
Page Overview
The Models page is divided into two main sections:
- Add Model: An expandable form for creating new model configurations.
- Models List: A list of existing models associated with the selected account and agent, providing details on status, provider, and capabilities.
Creating a New Model
To create a new model:
- Navigate to the Models page.
- Expand the "New Model" section.
- Fill out the configuration fields described below.
- Click Create Model.
Configuration Fields
Basic Information
- Name: A unique name for the model configuration (e.g.,
Gemini 2.0 Pro - Production). - Model Value: The specific identifier required by the provider (e.g.,
gemini-2.0-pro-exp-02-05,gpt-4o). - Status: Toggle to enable or disable the model.
- Agent(s): Associate this model configuration with specific agents. Select "All" to make it globally available within the account.
- Provider: The model service provider (e.g.,
google,openAI,anthropic,deepseek,meta,selected_agent_name,mistral). - Model Type: Defines the hosting environment (
cloud,local,ip,huggingface,custom).
Connection & Security
- Connector: Select a Connector that provides the necessary API keys or Service Account credentials for this model.
- Endpoint: The URL or path for local or IP-based models (e.g.,
http://localhost:11434for Ollama). - Port: The port number for IP-based models.
- Protocol: The communication protocol (
http,https,grpcwebsocket). - SSL Verify: Toggle whether to verify SSL certificates for custom endpoints.
- Version: Optional release or version string of the model.
Capabilities
- Capabilities: Multi-select of features supported by the model (
text,vision,audio,file,function-calling,reasoning). - Supports Vision/Audio/File: Explicit toggles for multimodal capabilities.
- Supports Function Calling: Enable if the model supports tool use.
- Supports Reasoning: Enable for models with dedicated reasoning chains (e.g., "Thinking" models).
- Supports Streaming: Enable if the model endpoint supports real-time token streaming.
- Supports JSON Mode: Enable if the model can be forced into strict JSON output mode.
Performance & Limits
- Max Tokens: Maximum context window supported by the model.
- Max Input/Output Tokens: Specific limits for prompt and response lengths.
- Rate Limits:
- RPS: Requests Per Second.
- RPM: Requests Per Minute.
- Daily: Maximum requests allowed in a 24-hour period.
- Timeout: Seconds to wait for a model response before failing.
- Retry Logic: Configure Retry Attempts and Backoff (ms) for handling transient errors.
Cost Tracking
- Input/Output Cost/Token: Decimal values representing the cost per token (e.g.,
0.0000025). - Cache Creation/Read Cost: Costs associated with context caching if supported.
Advanced Configuration
- Custom Headers (JSON): Additional HTTP headers required for custom API endpoints.
- Model Config JSON: Flexible storage for provider-specific configurations.
- Model Params JSON: Additional model parameters (e.g., default
top_p,top_k,repetition_penalty).
Listing and Managing Models
The Models List section displays all models associated with the selected criteria:
- Filtering: The list respects the agent selection in the sidebar.
- Details: Expanding a model entry reveals its full configuration, including connection details, costs, and limits.
- Edit: Click the edit icon to modify any aspect of the model configuration.
- Delete: Remove a model configuration from the account.
Usage Examples
Example 1: Google Gemini 3.1 Flash lite (Cloud)
Used for quick conversational agents and classification leveraging Google's infrastructure.
- Type:
cloud - Provider:
google - Model Value:
gemini-3.1-flash-lite - Connector:
My Google Cloud Connector
Example 2: Local Gemma4 3 via Ollama
Ideal for private processing on local hardware.
- Type:
local - Protocol:
ollama - Endpoint:
http://localhost:11434 - Model Value:
gemma4:12b
Example 3: Reasoning Model (GPT-o1)
Used for complex problem solving requiring deep thought chains.
- Provider:
openAI - Model Value:
o1-preview - Supports Reasoning:
True - Capabilities:
text,reasoning