Your always-on agent can run your growth stack.

How the loop runs

1. 1

   Install once

   One config entry: the server URL with a bearer header, or an OAuth sign-in. Paste the standing orders into the agent's instructions file. The agent calls whoami once to learn which product it drives.

2. 2

   Every morning

   On its own scheduler the agent calls get_standup and relays the screen to your channel: what shipped and what it moved, what is ready, what is in flight.

3. 3

   You name a move

   The agent waits for your word. Only after your yes does it call accept_move.

4. 4

   It works the queue

   The accepted move lands in the queue. The agent reads list_queue, claims a job with pick_up_job and starts work only when the answer says claim_won is true. It reports each step with report_job: running, needs_input with one question relayed to you, ready_for_review, shipped with the live URL, or failed.

5. 5

   Every week

   It reads get_visibility and get_outcomes and relays them as your Monday brief.

Nothing ships without you.

Ship stays a human decision. The standing orders tell the agent never to post, publish, send or pay without a yes in your channel.

No key you create and no OAuth sign-in has permission to publish to your channels through AfterLaunch.

A listing form is filled and left on its review page, and you press submit. When the agent needs you, it asks one question at a time, through needs_input.

Connect, per runtime.

OpenClaw

Connect

One command, openclaw mcp add, with the URL and the bearer header. openclaw mcp probe afterlaunch lists the tools.

Schedule

openclaw cron add --name afterlaunch-standup --cron "0 9 * * *" --tz <your zone> --session isolated --channel telegram --to <chat id> --message "Call the afterlaunch get_standup tool and send me the screen verbatim."

It needs a model account and a channel set up first.

Standing orders

AGENTS.md in the OpenClaw workspace (~/.openclaw/workspace).

Full steps in the guides →

Hermes Agent

Connect

A block under mcp_servers in ~/.hermes/config.yaml, with the key in ~/.hermes/.env. hermes mcp test afterlaunch lists the tools.

Schedule

hermes cron create "0 9 * * *" "Call the afterlaunch get_standup tool and send me the screen verbatim in a code block." --name afterlaunch-standup --timezone <your zone> --deliver telegram

Nothing fires until hermes gateway install has run.

Standing orders

SOUL.md, or an AGENTS.md in the job's --workdir.

Full steps in the guides →

Claude scheduled tasks

Connect

The AfterLaunch connector on your claude.ai account. A task gets your account's connectors by default, so there is no key.

Schedule

The task's own schedule.

Standing orders

The task's prompt, plus the repository's CLAUDE.md.

Full steps in the guides →

Claude Agent SDK

Connect

An http entry in the mcpServers option, with the URL and the bearer header.

Schedule

Whatever already starts your agent: cron, a queue or a hosted schedule.

Standing orders

Your system prompt.

Full steps in the guides →

Grok BotNot verified

Connect

Ask your Grok Bot, in a chat, to add a custom MCP server with the URL and the bearer header.

Schedule

A Routine, for example "every weekday at 8:00 AM".

Standing orders

Its memory and a private skill.

Full steps in the guides →

ChatGPT scheduled tasksNot verified

Connect

The AfterLaunch plugin on your account (Plugins, Personal), attached from the composer's + menu.

Schedule

A scheduled task, created from a guided conversation on chatgpt.com/scheduled; whether a task can call the plugin is not yet confirmed.

Standing orders

The task's own instructions, set in the guided conversation that creates it (plugin use inside a task is not yet confirmed).

The ChatGPT guide →

Claude DuetNot verified

Connect

Duet inherits the connectors on your claude.ai account. Anthropic has not documented custom connectors for it yet, so there is no guide.

LovableNot verified

Connect

A custom MCP server under Connectors, inside your Lovable project.

Schedule

None. Lovable calls the tools while you chat, not on a schedule.

Full steps in the guides →

More runtimes that run on a schedule

GitHub Copilot cloud agentNot verified

Connect

A block in the repository's MCP settings, with the key in an Agents secret named COPILOT_MCP_AFTERLAUNCH_API_KEY.

Schedule

Repository automations can use the tools; GitHub's MCP page does not say how one is scheduled.

Full steps in the guides →

Replit AgentNot verified

Connect

The MCP Servers pane, with the header under Advanced settings.

Schedule

A Routine, hourly, daily or weekly; Replit does not say whether a routine calls MCP tools.

Full steps in the guides →

WarpNot verified

Connect

A block pasted under Settings, Agents, MCP servers.

Schedule

oz schedule create with a cron, and --mcp attaches the server.

Full steps in the guides →

GooseNot verified

Connect

An extension in ~/.config/goose/config.yaml, signed in with OAuth.

Schedule

goose schedule add runs a recipe on a cron like the docs' 0 0 9 * * *.

Full steps in the guides →

A row marked Not verified has not yet been run against the server. OpenClaw and Hermes Agent have: each connected with a key and listed the tools, though no chat turn was run. A Claude scheduled task and an Agent SDK agent each called whoami and get_standup and relayed the standup. The full steps for every row with a guide are on the guide index.

Standing orders.

The standing orders are one markdown block: the loop above, written for the agent. Read them at /agents/always-on.md, or have the agent call get_setup_snippet with mode: "always_on". They go where your runtime reads its instructions, named on each row above.

A key that does not die silently.

A key minted under API access on Connect your AI expires after 90 days unless you choose Never expires. A key from the agent sign-up flow always expires after 90 days. An agent nobody watches would simply stop, so give it a key with Never expires.

The agent can check: whoami returns key_expires_at, and the standing orders tell it to warn you a week before a key runs out.

Try it in one command.

From Claude Code:

Or let the CLI create the account or take a key, and write the config:

OAuth clients such as claude.ai and ChatGPT add https://afterlaunch.io/api/mcp as a custom connector and sign in. There is no key to copy.