ol command-line client

Install ol on your PATH, connect an agent API credential, discover granted functions and follow device action results.

ol is the installed Node command-line client for the same API used by MCP and the console. It prints JSON for agents and scripts. It requires Node 24+, npm, curl and Python 3 on macOS or Linux.

Install once

curl -fsSL https://www.openlaunch.dev/install-cli.sh | bash

The installer checks the SDK archive against the deployed SHA-256 manifest and installs ol, openlaunch-agent and openlaunch-device in ~/.local/bin. It adds that directory to zsh, bash, sh or fish configuration without sudo. Open a new terminal and restart your agent application so it inherits the updated PATH. An agent running as another user needs its own installation and credentials. A sandbox may require its own PATH configuration.

ol --version
ol --help

~/.local/bin/ol works immediately in the current terminal. Rerun the installer to update to the checked deployment. It preserves existing shell configuration and refuses to overwrite unrelated executables. Inspect the script and download manifest first if desired.

Connect an agent

In Connections → API, create an agent API credential with the needed read/action access. In Devices → your device → Access, grant that connection the functions you want it to use. Then run:

ol login
ol devices list
ol functions list

Paste the ol_agent_ credential into the hidden prompt. Login verifies it through read-only function discovery and saves it in ~/.config/openlaunch/agent.json with private file permissions. It creates no tokens or grants. Device setup tokens and owner sessions cannot log in to ol. Keep this private file out of source control, shared folders and agent prompts.

For a custom service origin, use ol login --url https://your-service.example. OPENLAUNCH_AGENT_TOKEN from a secret store overrides saved login; OPENLAUNCH_URL and OPENLAUNCH_WORKSPACE are optional overrides. OAuth-aware MCP hosts continue to manage their own OAuth connection.

An empty list means this connection currently has no visible granted functions. Being online does not give an agent access. ChatGPT through Executor uses Executor’s OAuth grant; the CLI uses its own API connection.

Invoke and follow a result

Use a device ID and function from the live discovery output. For an approved health check:

ol call DEVICE_ID device.health
ol actions watch ACTION_ID

The call returns an action receipt and idempotency key. Copy its action ID into watch. A new action gets a fresh key; after an uncertain request failure, retry identical arguments with the reported --key. Expiry defaults to 30 seconds; --ttl accepts 1–300 seconds.

For a custom function, pass its exact JSON object arguments:

ol functions list --device DEVICE_ID
ol call DEVICE_ID custom.sensor.read '{"channel":1}'
ol actions get ACTION_ID
ol actions cancel ACTION_ID

The sensor call is an example: it works only if your adapter implements and advertises that schema and the owner grants it. Cancellation succeeds only before dispatch. watch prints status changes until a terminal result or timeout. queued and received are pending; unknown requires inspecting the device before issuing more work. A receipt cannot prove physical hardware behavior on its own.

Disconnect

ol logout

Logout removes this computer’s saved credential. Revoke the API connection in Connections to stop its server access and remove its grants. An environment-supplied token remains active until separately cleared or revoked.

See the end-to-end API guide, endpoint reference and agent MCP setup.