deploy-platform

Deploy the Platform

Coding-agent workflow: run as /deploy-platform or by describing the task.

Take the locally-proven platform to a live public URL. This creates real, billed cloud resources — say so before creating anything, and name the teardown script in the same breath.

Be self-driving: run every script and check yourself. Stop only for a provider login, a browser-only step, or a key only the user can mint — and tell them to run interactive commands in a separate terminal (CLI logins need a TTY and a browser). Never print secret values; the two exceptions are the JWT verification key (public) and the MCP_CONNECT_SECRET the up script prints itself.

Narrate: open with this map plus the cost sentence, then a line per step. On a redeploy (Step 3 finds the platform live) the map is three beats — push, prove, hand back — and the cost sentence is already spent.

Kicking off /deploy-platform. Here's the map:

1. Read the deploy layer — scripts + README, pick the mode
2. Preflight — provider CLI + login, cost and exit, production env
3. Deploy — the up script (compute + Postgres + public domain)
4. JWT key — connect os.agno.com Live, land the public key, sync it
5. Prove it live — logs, /docs 200, /mcp 401 challenge, UI Connect
6. Hand over — redeploy, logs, teardown, chat + coding-agent connect

This creates real, billed resources on your account. The exit is one
command away — the down script deletes everything (asks to confirm; --yes skips).

1. Read the deploy layer

Read AGENTS.md, the README's production section, and scripts/<provider>/ — never invent a step. Pick the mode by who provisions the target: Conduct (the up script provisions compute and URL on a managed provider — the main path), Conduct over owned infra (same scripts onto the user's own cluster; add Step 7), or Manual-guide (no deploy scripts; walk the README's deploy section with Steps 4, 5, and 7).

2. Preflight

• CLI + account: the provider CLI is installed and authed (a whoami-style probe). Not logged in → hand over the login command for a separate terminal, re-probe when they say ready.
• Cost + exit: one sentence — billed resources, down.sh deletes everything.
• Production env: .env.production (cp example.env .env.production if missing) with a real OPENAI_API_KEY set the setup-platform way (editor paste, never read or print). RUNTIME_ENV must not be dev there — it syncs to the cloud and disables auth.
• Unattended-run inputs: for each provisioning call the up script makes, does it pin the account scope (workspace, org, project, region)? If the account has more than one and nothing pins it, the CLI opens a picker that fails or hangs — stop before creating anything and hand the user the pinning flag/env var or the one interactive init/link command. Never let it resolve by guess.