Model Parameters Configuration

CronAI supports model-specific parameters that allow you to fine-tune AI model behavior for each prompt. This document explains how to configure and use these parameters.

Supported Parameters

Common Parameters

The following parameters are supported across all models:

Parameter	Type	Range	Description
temperature	float	0.0 - 1.0	Controls response randomness (higher = more random)
max_tokens	int	> 0	Maximum number of tokens to generate
model	string	-	Specific model version to use

Model-Specific Parameters

Each model can also be configured with specific parameters using the prefix notation model_name.parameter:

OpenAI

Parameter	Type	Description
openai.model	string	Specific OpenAI model to use

Claude

Parameter	Type	Description
claude.model	string	Specific Claude model to use

Gemini

Parameter	Type	Description
gemini.model	string	Specific Gemini model to use

Model-Specific Default Values

OpenAI Default Settings

• Default Model: gpt-3.5-turbo
• Supported Models:
  • gpt-3.5-turbo - Fast and cost-effective for most tasks
  • gpt-4 - Strong reasoning and instruction following

Claude Default Settings (v0.0.2+)

• Default Model: claude-3-5-sonnet-latest

• Supported Models:

  Claude 4 Models (Latest - v0.0.2+):

  • claude-4-opus-latest / claude-4-opus-20250514 - Most capable model for complex tasks
  • claude-4-sonnet-latest - Balanced performance and cost
  • claude-4-haiku-latest - Fastest and most efficient

  Claude 3.5 Models (v0.0.2+):

  • claude-3-5-opus-latest / claude-3-5-opus-20250120 - Most capable 3.5 model
  • claude-3-5-sonnet-latest / claude-3-5-sonnet-20241022 / claude-3-5-sonnet-20240620 - Balanced 3.5 model
  • claude-3-5-haiku-latest / claude-3-5-haiku-20241022 - Fast 3.5 model

  Claude 3 Models (v0.0.2+):

  • claude-3-opus-latest / claude-3-opus-20240229 - Most powerful Claude 3 model
  • claude-3-sonnet-latest / claude-3-sonnet-20240229 - Balanced performance and speed
  • claude-3-haiku-latest / claude-3-haiku-20240307 - Fast and economical

• Model Aliases (v0.0.2+):

  • opus → claude-4-opus-latest
  • sonnet → claude-4-sonnet-latest
  • haiku → claude-4-haiku-latest
  • 3.5-opus → claude-3-5-opus-latest
  • 3.5-sonnet → claude-3-5-sonnet-latest
  • 3.5-haiku → claude-3-5-haiku-latest
  • 3-opus → claude-3-opus-latest
  • 3-sonnet → claude-3-sonnet-latest
  • 3-haiku → claude-3-haiku-latest

Gemini Default Settings

• Default Model: gemini-pro
• Supported Models:
  • gemini-pro - Original Gemini model

SDK Implementation

CronAI uses official client SDKs for all supported AI models:

• OpenAI: Uses the official github.com/sashabaranov/go-openai SDK
• Claude: Uses the official github.com/anthropics/anthropic-sdk-go SDK
• Gemini: Uses the official github.com/google/generative-ai-go SDK

Configuration Methods

You can configure model parameters in three ways, listed in order of precedence:

1. Task-specific parameters in the config file
2. Environment variables
3. Default values

1. Task-specific Configuration

In the cronai.config file, you can specify model parameters using the prefix model_params::

# Format: timestamp model prompt response_processor [variables] [model_params:...]
0 8 * * * claude product_manager file-output.txt model_params:temperature=0.8,model=claude-3-opus-20240229

You can also include both variables and model parameters:

0 9 * * 1 openai report_template github-issue:owner/repo reportType=Weekly,date={{CURRENT_DATE}} model_params:temperature=0.5,max_tokens=4000,model=gpt-4

Using Model-Specific Parameters

For model-specific configuration, use the prefix notation:

# Use OpenAI-specific parameters
0 9 * * 1 openai report_template file-output.txt model_params:openai.model=gpt-4

# Use Claude-specific parameters
0 8 * * * claude product_manager file-output.txt model_params:claude.model=claude-3-opus-20240229

# Use Claude 4 models (v0.0.2+) - using aliases
0 8 * * * claude product_manager file-output.txt model_params:claude.model=opus
0 8 * * * claude product_manager file-output.txt model_params:claude.model=claude-4-opus-latest

# Use Claude 3.5 models (v0.0.2+)
0 8 * * * claude product_manager file-output.txt model_params:claude.model=3.5-sonnet

# Use Gemini-specific parameters
*/15 * * * * gemini system_health file-output.txt model_params:gemini.model=gemini-pro

2. Environment Variables

You can set global defaults for all tasks using environment variables:

# Common parameters
MODEL_TEMPERATURE=0.7
MODEL_MAX_TOKENS=2048

# Model-specific parameters
OPENAI_MODEL=gpt-4
CLAUDE_MODEL=claude-3-opus-20240229
GEMINI_MODEL=gemini-pro

3. Command Line Parameters

When using the run command, you can specify model parameters with the --model-params flag:

# Using common parameters
cronai run --model openai --prompt weekly_report --processor file-output.txt --model-params "temperature=0.5,max_tokens=4000,model=gpt-4"

# Using model-specific parameters
cronai run --model gemini --prompt system_health --processor file-output.txt --model-params "gemini.model=gemini-pro"

Examples

Low Temperature for Consistent Output

# Run system health check with very precise (low temperature) settings
*/15 * * * * claude system_health file-health.log cluster=Primary model_params:temperature=0.1,max_tokens=1000

Specific Model Version

# Run weekly with OpenAI using a specific model
0 9 * * 1 openai report_template file-report.log model_params:openai.model=gpt-4

GitHub Processor Example

# Create a GitHub issue with the weekly report
0 9 * * 1 claude weekly_report github-owner/repo reportType=Weekly,date={{CURRENT_DATE}} model_params:temperature=0.7

Console Output Example

# Output system health check to console (useful for testing)
*/30 * * * * gemini system_health console model_params:temperature=0.3,max_tokens=2000

Integration with Variables

Model parameters can be used alongside variables:

0 9 * * 1 openai report_template file-report.log reportType=Weekly,date={{CURRENT_DATE}} model_params:temperature=0.5,max_tokens=4000,openai.model=gpt-4

Advanced Configuration

Custom API Endpoints

For organizations using proxy services or custom endpoints for AI models, CronAI supports custom base URLs through environment variables:

• OpenAI: Set OPENAI_BASE_URL environment variable to point to a custom endpoint
• Claude: Set ANTHROPIC_BASE_URL environment variable to point to a custom endpoint

Timeout Configuration

All model clients have a default timeout of 120 seconds (2 minutes) for API requests. This can be adjusted by setting the appropriate environment variables:

• OpenAI: Set OPENAI_TIMEOUT to the desired timeout in seconds
• Claude: Set ANTHROPIC_TIMEOUT to the desired timeout in seconds
• Gemini: Set GEMINI_TIMEOUT to the desired timeout in seconds

Post-MVP Features

The following features are planned for post-MVP releases:

• Advanced parameters like top_p, frequency_penalty, and presence_penalty
• System message customization
• Model fallback mechanism for automatic model switching on failure
• Safety setting configurations for Gemini
• Detailed error handling with retry mechanisms