AfterLaunch 1.1: your always-on agent does the growth work

Always-on agents reached ordinary users this autumn: Grok Bot and ChatGPT dots run on a schedule without you. AfterLaunch 1.1 gives that agent one job, your organic growth. On 12 Nov 2026 we connected a Grok Bot, gave it our standing orders and said nothing else. It checked the top-ranked move against the live site, found the page already existed, brought the next move as one decision, and after a yes came back with a draft.

Organic growth here means two channels: being found on Google, and being named when someone asks an AI search engine for a tool like yours. AfterLaunch reads what ChatGPT, Gemini, Perplexity and Google AI Overviews say about your product, ranks the next growth move and briefs it with its evidence. Each morning the agent reads that record and brings you one decision. After your yes it does the work and brings you the draft. Nothing ships until you say yes.

What happened on Grok Bot

We connected Grok Bot through its own custom connector form: Connect apps, Add a custom connector, Remote HTTPS, with the address and one header carrying an AfterLaunch key. It called whoami and named the product. We pasted the standing orders into a chat, and it answered:

I've saved the orders as a skill.

Then it started on its own. Before bringing anything it checked the top-ranked move against the live site. The quotes below are its words, with the rival and the page paths renamed:

The top-ranked move asks for a Vantico alternative page, but your site already has /learn/vantico-alternatives and /vs/vantico, so I'm checking move 2 before I bring you anything.

Its first message introduced itself in four lines, said what it never does without a yes, and brought one decision with its evidence:

I never post, publish, send, pay, or start a paid scan or loop without a yes from you here.

After "yes to both" it set its own routine, every day around 8:00 AM with the Monday review folded in, marked the already-live move done, claimed the next one and came back with a draft comparison article. It could not stage the draft, because it had no access to the site’s repository. It said so and asked: stage it, change it, or drop it. We have not yet seen a second morning.

What the agent does on each run

Every run starts with whoami, which names the product the key drives and when the key expires. Each morning the agent reads the standup, the scoreboard, the queue and what is new in your growth moves, then sends you one short message with one decision on top: something only you can unblock, a new scan, a result that moved, or the top-ranked move. Before it brings a move it checks the work is not already done, already live or asked for twice, and offers a duplicate once as a clean-up. When nothing changed, it sends a line or nothing. Those reads run on your agent’s own model, so the cost of the morning check is yours.

When you say yes in the agent’s channel, it works the move through the queue:

• Accepts the move with accept_move, which puts the work in the queue.
• Claims the job with pick_up_job, and starts only when the answer says claim_won is true.
• Reads the brief with get_move: the evidence, the deliverable and the bar the finished work has to clear.
• Asks at most one question, as needs_input, and waits for your answer in the same channel.
• Reports each step with report_job: running, ready_for_review with the draft, shipped with the live address, or failed with the reason.

You can let it do more, one rung at a time, in the same channel. Under draft ahead the morning message is the finished draft, waiting on stage it, change it or drop it. Under stage ahead, which you grant per service, the orders tell it to load each finished draft where it will be published, one press short, on the services you granted. One word revokes either.

On Mondays it sends one review, each item ending in a yes or no: what moved since the last review, the latest scan and when the next is due, questions worth tracking or dropping, and every open loop. AfterLaunch re-scans on its own schedule, so the agent reports results rather than buying new scans. The tools that cost money, refresh_scan and replenish_feed among them, run only after your yes.

Tap to zoom

Nothing ships until you say yes

The standing orders forbid the agent to post, publish, send, pay or submit anything without a yes from you in its channel. They tell it to stage work one press short where it can reach: a post loaded in the composer, an email saved as a draft, a listing form filled in and left on its review page. You press the last button. The agent reports a move shipped only after your yes, with the live address, once it can see the page live.

AfterLaunch records the work and keeps the history. It never publishes on its own, and shipping stays your decision.

Which agents, and what we have run

• Grok Bot: the full loop, run for real on 12 Nov 2026, from connection to a draft.
• ChatGPT dots: a dot uses the plugins installed on your ChatGPT account, so it reaches AfterLaunch through the AfterLaunch plugin, which we have run in a real ChatGPT account.
• Claude scheduled tasks: a task with the AfterLaunch connector called whoami and get_standup and relayed the standup on 7 Nov 2026, under the earlier orders.
• For developers: a Claude Agent SDK agent did the same on 7 Nov 2026, and fresh OpenClaw and Hermes Agent installs connected with a key and listed the tools. No chat turn was run on those two.

Coming soon to Claude Duet. Today, use a Claude scheduled task.

Every other client’s set-up is on the guide index, row by row. Grok Bot is a product of xAI, ChatGPT of OpenAI and Claude of Anthropic. AfterLaunch is independent and is not affiliated with or endorsed by them.

Set it up

Start with the free scan, which creates your account and your first growth moves. Where an agent takes a key, create one under API access on Connect your AI and choose Never expires, so an agent nobody is watching does not stop silently. The standing orders tell the agent to warn you seven days before a key runs out.

Grok Bot

• Create a key in AfterLaunch under API access on Connect your AI, set to Never expires, and keep it for Grok Bot alone.
• In your Grok Bot app open Connect apps, choose Add a custom connector, then Remote HTTPS, and name it AfterLaunch.
• Paste the server URL: https://afterlaunch.io/api/mcp
• Add one header: the name Authorization, and as its value the word Bearer, a space, then your key. Press Add plugin.
• Ask it to call the AfterLaunch whoami tool. It names your product.

Then paste the standing orders into a chat and ask it to save them as a skill. Its home for them is a skill in your skill library (paste the orders and ask it to save them as a skill).

ChatGPT dots

Add the AfterLaunch plugin to ChatGPT, using the steps on the ChatGPT guide, and sign in. Your dot uses the plugins installed on your ChatGPT account: tell it to use AfterLaunch, give it the standing orders as custom rules, and ask it for a morning task.

A Claude scheduled task

A scheduled task uses the AfterLaunch connector on your claude.ai account, so there’s no key to expire while nobody is watching.

• Connect AfterLaunch to your Claude account once, as a custom connector: on claude.ai open Customize, then Connectors, add a custom connector with the server URL https://afterlaunch.io/api/mcp, then press Connect and approve AfterLaunch.
• Create a scheduled task at claude.ai/code/routines, or with /schedule in Claude Code. A task gets your account's connectors by default, so AfterLaunch is there without a key.
• Put the standing orders from /agents/always-on.md in the task's prompt.
• Run it once and read its first message: it introduces itself and brings one decision.

In a terminal: OpenClaw

One command adds the server. Replace YOUR_AFTERLAUNCH_API_KEY with your key.

OpenClaw probes the server before it saves, and refuses a wrong key. openclaw mcp probe afterlaunch lists the tools. The standing orders go in AGENTS.md in the OpenClaw workspace (~/.openclaw/workspace). The morning run is an openclaw cron add job. Run openclaw onboard first: it sets up the gateway, a model account and a channel, and OpenClaw refuses the job until it has run.

In a terminal: Hermes Agent

Add this block to ~/.hermes/config.yaml, under your existing mcp_servers key if you have one. Put AFTERLAUNCH_API_KEY=... in ~/.hermes/.env, which Hermes reads first.

hermes mcp test afterlaunch connects and lists the tools. The standing orders go in SOUL.md, or an AGENTS.md in the job's --workdir. They also install as a skill, described below. The morning run is a hermes cron create job, and nothing fires until hermes gateway install has run.

In code: the Claude Agent SDK

Pass this block as the mcpServers option of query(). In code, ${AFTERLAUNCH_API_KEY} stands for process.env.AFTERLAUNCH_API_KEY, so the key stays in the environment and out of your source.

With the variable empty, the run still ends as a success and the server is marked failed in the system init message, so check that the afterlaunch entry there reads connected. The standing orders go in your system prompt. Your agent decides when it runs: cron, a queue or a hosted schedule.

What changed in 1.1

The hosted MCP server lives at https://afterlaunch.io/api/mcp. It takes your key in an Authorization: Bearer header, or a sign-in where a client connects through a connector. No tool was added for this release. Version 1.1 adds what an agent needs when nobody is at the keyboard:

• A connect line for each always-on agent, kept in the same client table every guide on the site renders from.
• Standing orders written for a run nobody watches, served at /agents/always-on.md and by the get_setup_snippet tool with mode always_on.
• The same orders as an installable Agent Skill.
• The key’s expiry date in every whoami answer, so the agent can warn you before its key stops working.

The installable skill

The standing orders also ship as an Agent Skill. Everything after the skill’s frontmatter is the same text as /agents/always-on.md, so a pasted copy and an installed copy say the same thing. Hermes Agent installs it with one command, and you answer y at the prompt:

How AfterLaunch works with your agent, step by step, is on its own page, with a blank template of the same structure for your own tools at the bottom.

What it costs to try

The free scan costs nothing and asks for no card. A 14-day trial follows, also with no card, and connecting your own agent is part of it. The scan and the trial put your customers’ questions to ChatGPT and Perplexity, and during the trial they are asked every week.

At the end of day 14 the weekly checks stop and no new moves arrive. Founder, the one paid plan, keeps them running and adds Gemini and Google AI Overviews, at $149 a month.

The agent runs on your own subscription. AfterLaunch does not charge per run.

What this does not do

• It doesn’t publish without you. AfterLaunch publishes nothing itself, and the agent publishes only after your yes in its channel.
• It does not promise rankings or mentions. Google and the AI engines decide what they show. AfterLaunch asks the same questions each week and keeps the answers, so you can see what changed after a move shipped.
• It does not spend on its own. The tools that cost money run only after your yes, and the agent reports the scans AfterLaunch already runs rather than buying new ones.