CLI Reference¶
OpenSage-ADK provides a command-line interface for interactive development and utility tasks.
Commands¶
| Command | Description | Reference |
|---|---|---|
opensage |
Main entry point | Run uv run opensage --help |
opensage web |
Interactive web UI for agent development | Run uv run opensage web --help |
opensage run |
Terminal agent runner without the web UI | Run uv run opensage run --help |
opensage review |
Read-only session viewer for saved trajectories | Run uv run opensage review --help |
opensage dependency-check |
Verify external dependencies | Run uv run opensage dependency-check --help |
Usage: opensage [OPTIONS] COMMAND [ARGS]...
OpenSage CLI tools.
Options:
--help Show this message and exit.
Commands:
dependency-check Check OpenSage external dependencies.
review Read-only session viewer — no LLM, no sandbox, no core...
run Run an agent interactively in the terminal (no web UI).
web Starts an OpenSage-flavored Web UI: prepare...
opensage run¶
Runs an agent from the terminal instead of starting the web UI.
| Flag | Purpose |
|---|---|
--config |
Path to the OpenSage TOML config |
--agent |
Path to the agent folder |
--log_level |
Logging level: debug, info, warning, error, or critical |
--auto_cleanup |
Whether to clean up sandboxes on process exit |
--resume |
Resume the most recently saved session |
--resume-from |
Resume a specific saved session |
--prompt |
Send one prompt, run to completion, then exit |
opensage review¶
Opens a read-only Web UI for a saved session. It does not start LLM calls, sandboxes, or core runtime services.
| Argument / Flag | Purpose |
|---|---|
SESSION_PATH |
Optional saved session directory, directory name, or suffix. Defaults to the most recent saved session |
--host |
Binding host for the viewer |
--port |
Port for the viewer |
--log_level |
Logging level |
--live |
Poll traj.json files for updates while watching a running opensage run session |
opensage web vs Evaluations¶
| Aspect | opensage web |
Evaluations |
|---|---|---|
| Use Case | Development, debugging | Performance measurement |
| Interaction | Interactive chat | Batch processing |
| Sessions | Single long-lived | Multiple short-lived |
| Parallelism | Single user | Multiple tasks |
| Output | Real-time events | Saved results files |
See also: Evaluations