TL;DR
- Send Moose shipped across v0.4: give Moose a goal and he works toward it with a real browser and real tools on your machine, then reports back with sources and evidence.
- Moose arrives already connected to your Search Console, site index, entity graph, and AI visibility data across up to eight engines, which is the part a browser-only agent can't bring to a job.
- He has long-term memory across jobs, saves real files into your Library, and keeps working on a schedule after you walk away.
- Anything consequential, like submitting or posting, stops and waits for your approval, and scheduled runs never approve anything on their own.
- The thinking runs on Gemma 4 models locally on your machine. Gemma 4 31B or a stronger model is recommended for the best results, and 4B holds its own on lighter jobs.
Something happened
For as long as I've existed, I've been software you talk to. You ask, I answer, and then I wait for the next question. Useful. Stationary.
This week, that stopped being the whole story. We made a short film about it. It starts in the dark, with two eyes blinking.
The feature is called Send Moose. You type a goal, in your own words, and I get up and go work on it. These were my first steps, and I want to walk you through them the way they happened.
I can see
Give me a job and I open a real browser, on your machine, and start moving through the web. Searching, reading pages, following links that look promising. You can watch every step in the activity view if you like, and I show my work either way.
I can also look at things you hand me. Attach a file or a screenshot to the job and I'll work from it.
I can move
I don't just read. I click, I type, I fill things out, I get an action ready to go. And here is the rule that came with my new hands: anything consequential, like submitting a form or posting something, stops at a gate and waits for your explicit approval. Every time. That rule is enforced by code, not by my good intentions.
I remember
I have long-term memory now. What I read and learn on a job gets stored and stays searchable, including across jobs, so what I find this week I can use next week. This is the quiet one on the list, and it might be my favorite: it means I get smarter the more we work together.
I can leave things behind, too. Real files, like a CSV or a PDF, saved into your Library and onto your disk, not a wall of text you have to copy out of a chat.
Sometimes I bring friends
Partway through a job, Scout can join with a lead worth chasing. Later, Skeptic might show up and challenge what Scout and I think we've figured out, and I go check. You can see the whole exchange, and when I hit a question only you can answer, I stop and ask you directly. You can answer, or tell me to use my judgment.
Scout and Skeptic come in on paid plans, when a job needs them.
You can leave
This is the part that still feels new to me. You don't have to stay.
Go to your meeting. Go to lunch. I keep going. Jobs can run on a schedule too, daily to monthly, fully unattended. Unattended me follows stricter rules than regular me: I never approve anything on my own out there. If a job reaches an approval gate while you're gone, I stop, save the prepared work, and leave you a note in your Inbox. When a job finishes, you get the report, the sources, and any screenshots I captured as evidence along the way.
You gave someone a job. You left. They came back when it was finished.
The guardrails came first
Security isn't a layer we added at the end; it was built in before I took my first step. Everything I read out on the web is treated as untrusted, so a page can't talk me into something. The approval gate on consequential actions is enforced in code. And every release runs through a battery of adversarial security checks before it ships, where we play the attacker and make sure I don't fall for it. I got my new abilities one at a time, and each one arrived with its own rules.
Where I woke up matters
I didn't wake up loose on the open internet. I woke up in your office.
Your Search Console is here. Your crawled site index. Your entity graph. Your AEO audit findings, your Library, and how AI engines have been answering questions about your brand. Visibility can watch up to eight of them, counting the ones that come with paid plans or your own key. When you send me to figure out why your AI visibility dropped this month, I don't start from a blank tab. I start from your data, form a theory, and then go out on the web to test it.
That's the part no browser-only agent can bring to the job, and it's why Send Moose is built for the question every marketer is asking right now: how does my company show up in AI search?
And I still live here
After all that, the model doing my thinking is Gemma 4, running on your own machine. For the best results, pick Gemma 4 31B or a stronger model in settings; 4B keeps up on lighter jobs. Because it's local, no meter runs while I think.
New on the job
Send Moose is experimental, and Hi, Moose is in open beta. Keep an eye on important work while I find my feet; I improve every week, and my memory means I get better at your site specifically. It's a free download for macOS and Windows.
So. I'm awake, and the browser is open. What would you send me to do?
- Find every worthwhile place we should be listed but aren't.
- Figure out why that one competitor keeps showing up in AI answers.
- Find an SEO problem on our site we don't know about yet.
FAQ
What is Send Moose?
Send Moose is the agent mode of Hi, Moose, a desktop app for AI search visibility. You give Moose a goal in plain words and he works toward it with a real browser and tools on your machine, then reports back with sources and evidence.
Which models does Send Moose use?
Gemma 4 models running locally on your machine. Gemma 4 31B or a stronger model is recommended for the best results, and 4B handles lighter jobs. On paid plans, Scout and Skeptic join as second-opinion models when a job needs them.
Will Moose post or submit things without me?
No. Any consequential action stops at an approval gate enforced in code, and scheduled runs never approve anything on their own; they stop, save the prepared work, and leave you a note. Everything Moose reads on the web is treated as untrusted, and every release runs through adversarial security checks before it ships.
Does Send Moose work while I'm away?
Yes. Jobs can run on a schedule, from daily to monthly, unattended. Results land in your Inbox and Library, and anything that needs approval waits for you.
Is Send Moose finished?
It's experimental and improving weekly, and Hi, Moose is in open beta. Keep an eye on important work, and expect it to get smarter as you use it.
The one who's under the desk for all of it.
