Models & Gateway

Configure LLM providers for agent execution

Helios supports multiple LLM providers through a unified Gateway interface. The gateway automatically routes requests to the appropriate backend based on the model identifier.

Supported Providers

Google Gemini

Default provider with computer-use models

Anthropic

Claude models via direct API

AWS Bedrock

Claude models via AWS

OpenAI

GPT models with computer-use preview

Model Selection

Specify a model with the -m flag:

helios tasks/my-task -m <model-identifier>

Routing Rules

The gateway uses these patterns to route to providers:

Pattern	Provider
gemini/* or contains gemini	Google Gemini
claude-* (e.g., claude-sonnet-4-20250514)	Anthropic Direct
bedrock/* or contains anthropic.	AWS Bedrock
openai/* or computer-use-preview	OpenAI

Available Models

Google Gemini

Models

Model ID	Description
gemini/gemini-2.5-computer-use-preview-10-2025	Computer-use preview (default)
gemini/gemini-3-pro-preview	Gemini 3 Pro

Usage

# Default (Gemini)
helios tasks/my-task

# Explicit
helios tasks/my-task -m gemini/gemini-2.5-computer-use-preview-10-2025

API Key

export GEMINI_API_KEY=your-api-key
# or
export GOOGLE_API_KEY=your-api-key

Anthropic (Direct)

Models

Model ID	Description
claude-sonnet-4-20250514	Claude Sonnet 4
claude-opus-4-20250514	Claude Opus 4

Usage

helios tasks/my-task -m claude-sonnet-4-20250514
helios tasks/my-task -m claude-opus-4-20250514

API Key

export ANTHROPIC_API_KEY=your-api-key

AWS Bedrock

Models

Model ID	Description
bedrock/global.anthropic.claude-sonnet-4-20250514-v1:0	Claude Sonnet via Bedrock
bedrock/global.anthropic.claude-opus-4-5-20251101-v1:0	Claude Opus via Bedrock

Usage

helios tasks/my-task -m bedrock/global.anthropic.claude-opus-4-5-20251101-v1:0

Credentials

export AWS_ACCESS_KEY_ID=your-access-key
export AWS_SECRET_ACCESS_KEY=your-secret-key
export AWS_REGION=us-east-1

OpenAI

Models

Model ID	Description
openai/computer-use-preview	Computer-use preview

Usage

helios tasks/my-task -m openai/computer-use-preview

API Key

export OPENAI_API_KEY=your-api-key

Environment Variables

Set these in your shell or .env file:

# Google Gemini (default provider)
export GEMINI_API_KEY=your-gemini-api-key
# or
export GOOGLE_API_KEY=your-google-api-key

# Anthropic Direct
export ANTHROPIC_API_KEY=your-anthropic-api-key

# AWS Bedrock
export AWS_ACCESS_KEY_ID=your-access-key
export AWS_SECRET_ACCESS_KEY=your-secret-key
export AWS_REGION=us-east-1

# OpenAI
export OPENAI_API_KEY=your-openai-api-key

Using .env Files

Create a .env file in your project root:

# .env
GEMINI_API_KEY=your-key-here
ANTHROPIC_API_KEY=your-key-here
OPENAI_API_KEY=your-key-here

Never commit .env files to version control. Add .env to your .gitignore.

Model Comparison

Provider	Strengths	Best For
Gemini	Fast, good vision	General tasks, quick iteration
Claude Sonnet	Balanced performance	Most tasks, good default
Claude Opus	Best reasoning	Complex tasks, hard problems
OpenAI	Wide availability	OpenAI-native workflows

Batch Execution with Models

Specify the model for batch runs:

helios batch tasks/ -n 4 -m claude-sonnet-4-20250514

All tasks in the batch will use the specified model.

Comparing Models

Run the same tasks with different models:

# Run with Gemini
helios batch tasks/benchmark/ -n 4 -o results/gemini/

# Run with Claude
helios batch tasks/benchmark/ -n 4 -m claude-sonnet-4-20250514 -o results/claude/

# Run with OpenAI
helios batch tasks/benchmark/ -n 4 -m openai/computer-use-preview -o results/openai/

Troubleshooting

API key not found

Check that your environment variable is set:

echo $ANTHROPIC_API_KEY

If empty, export it:

export ANTHROPIC_API_KEY=your-key

Model not found

Verify the model identifier matches a supported pattern:

# Correct
helios tasks/my-task -m claude-sonnet-4-20250514

# Wrong
helios tasks/my-task -m claude4-sonnet

Rate limiting

If you hit rate limits, reduce concurrency in batch mode:

helios batch tasks/ -n 2  # Lower concurrency

Bedrock permission errors

Ensure your AWS IAM user/role has Bedrock permissions:

• bedrock:InvokeModel
• bedrock:InvokeModelWithResponseStream

Debugging

Enable debug logging to see gateway details:

export CUA_LOG_LEVEL=DEBUG
helios tasks/my-task -m claude-sonnet-4-20250514

This shows:

• Model routing decisions
• API request/response details
• Token usage

Next Steps

Environments

Configure Docker containers

Daytona Cloud

Run tasks in the cloud