helios CLI is the primary interface for running tasks.
Commands
helios (run)
Run a single task.| Argument | Description |
|---|---|
TASK_PATH | Path to the task directory |
| Option | Short | Default | Description |
|---|---|---|---|
--watch | false | Start web viewer at localhost:8080 | |
--port | -p | 8080 | Port to run the web viewer on (when using —watch) |
--model | -m | Gemini | Model identifier |
--interactive | -i | false | Enable pause/resume with ‘p’ key |
--n-attempts | -k | 1 | Number of attempts (for pass@k evaluation) |
--output | -o | output | Output directory |
--provider | docker | Environment: docker or daytona |
helios batch
Run multiple tasks in parallel.| Argument | Description |
|---|---|
DIRECTORY | Directory containing tasks |
| Option | Short | Default | Description |
|---|---|---|---|
--concurrent | -n | 2 | Number of concurrent tasks |
--n-attempts | -k | 1 | Number of attempts per task (for pass@k) |
--model | -m | Gemini | Model identifier |
--output | -o | output | Output directory |
--quiet | -q | false | Show only aggregate progress |
--pattern | -p | **/task.toml | Glob pattern for finding tasks |
--provider | docker | Environment: docker or daytona |
helios dev
Run in development mode with mock data.Environment Variables
LLM Providers
Cloud Providers
Debugging
Model Identifiers
| Provider | Model ID |
|---|---|
| Gemini | gemini/gemini-2.5-computer-use-preview-10-2025 |
| Anthropic | claude-sonnet-4-20250514 |
| Anthropic | claude-opus-4-20250514 |
| Bedrock | bedrock/global.anthropic.claude-sonnet-4-20250514-v1:0 |
| Bedrock | bedrock/global.anthropic.claude-opus-4-5-20251101-v1:0 |
| OpenAI | openai/computer-use-preview |
Output Structure
Single Task
Batch
-k runs, each task folder includes attempt_001/, attempt_002/, etc.
Exit Codes
| Code | Meaning |
|---|---|
0 | Success (all tasks passed) |
1 | Failure (one or more tasks failed) |
2 | Error (execution error) |
Tips
Use --watch for debugging
Use --watch for debugging
The web viewer shows real-time execution. Essential for GUI tasks.
Start with low concurrency
Start with low concurrency
Begin with
-n 2 and increase based on system resources.Use interactive mode for investigation
Use interactive mode for investigation
Press ‘p’ to pause and inspect agent state with
-i.Organize outputs by date
Organize outputs by date