Quickstart

Zero to an agent-run action in about five minutes. Steps 1–5 are performed by a human (browser + terminal); step 6 belongs to the coding agent.

1. Sign up

Create an account at https://omnicli.io/sign-up and verify your email — an unverified account can't approve devices or serve definitions.

2. Build a definition

A definition is the catalog the CLI will carry: the platforms and actions picked in the dashboard. Choose only what the workflow needs — the agent's context stays small, and so does its blast radius. See Definitions.

3. Install the CLI

The dashboard provides a copy-paste install one-liner bound to the definition (the token in it is scoped to that build, expires, and can be revoked):

curl -fsSL https://omnicli.io/install.sh | bash -s -- <INSTALL_TOKEN>

It downloads the signed binary, verifies checksums, and installs omni-cli. Windows has a PowerShell equivalent. Details in Install & Sign In.

4. Sign the device in (human-only)

omni-cli account login

The CLI prints enter code: XXXXXXXX and a verification URL. A human opens it, approves the device, and the CLI stores a device-bound token. This step is deliberately interactive — an agent that tries it gets exit code 6 (interactive_required).

5. Connect a platform (human-only)

omni-cli github login --context work

The CLI prompts for a token (some platforms also ask for an email or host — see Platform Logins). Credentials are stored locally and never sent to OmniCLI's servers. Contexts explains work vs personal.

6. The agent discovers and runs

omni-cli discovery                     # platform index (JSON)
omni-cli discovery github              # github's actions
omni-cli discovery github repo.get     # one action in full
omni-cli github repo get --owner octocat --repo hello-world --json

Success writes the platform's response to stdout and exits 0; failures exit non-zero with a structured error on stderr. The drop-in agent instructions live in Connect Your Agent.

Read this page as markdown — the copy your coding agent fetches.