How I Built a Personal Assistant Using ChatGPT-6 Astra (AI Agents) — Transcript
Full transcript
- 0:00What a GDT6 Aster could work for you
- 0:0224/7. And not just when you open a chat,
- 0:05but around the clock, even while you are
- 0:08fast asleep, that's exactly what I
- 0:10built. My assistant agent runs on AA on
- 0:13its own computer in the cloud, and at
- 0:156:00 a.m. in the morning before I'm even
- 0:18awake, it has already checked my
- 0:20calendar, researched everyone that I'm
- 0:22going to be meeting, and also drafted
- 0:24email replies for me. Then during the
- 0:26day, it keeps watch and it flags
- 0:29anything urgent the moment it lands. And
- 0:32it doesn't work alone. It literally
- 0:34manages its own team of agents that do
- 0:37the work for it. So in today's video,
- 0:40I'll show you how to build your own
- 0:42assistant agent with memory tools and a
- 0:45schedule running on Aster. Then I'm
- 0:48going to turn it into a manager that
- 0:51runs its own team of AI agents. Now,
- 0:54before we jump in, every single prompt
- 0:56and the exact setup from today's video
- 0:58is in our free school community. The
- 0:59link is down in the description below.
- 1:01Right now, let's hop over to the laptop
- 1:04and let's get started.
- 1:10Now, just a quick question that you're
- 1:11probably asking, why not just build this
- 1:14inside chatbt or claude? Well, those
- 1:17apps are those apps are great and they
- 1:20are adding more agent features every
- 1:22single month. But Hyper Agent is built
- 1:25for one thing, agents that can act like
- 1:29real employees. Every agent gets its own
- 1:33computer in the cloud and its own
- 1:35browser. It runs on a schedule whether
- 1:37my laptop is open or not, and it
- 1:40remembers what it learns. That means
- 1:42that I can pick a different model for
- 1:44every agent. And the big one for today,
- 1:47agents can manage other agents and my
- 1:51whole team can use them. The link to
- 1:54Hyper Agent is down in the description
- 1:55below. So go and sign up right now and
- 1:58let's jump in. So Aster is the brain.
- 2:01Okay, Hyper Agent is the system that it
- 2:04works in. Now let's go into more detail
- 2:06about all of that.
- 2:11Now, we are about to start with the
- 2:13build. But before we get into that,
- 2:15here's about 30 seconds that I'm just
- 2:18going to do. It's just a 30-cond tour so
- 2:20that you know where everything is. Okay.
- 2:22So, up top there's new thread is where
- 2:25you are going to give any agent a task.
- 2:28Inbox is where your agents come to you
- 2:30with questions, approvals, and anything
- 2:32that needs your attention. Now then
- 2:34there's also resources, skills or
- 2:36workflows that your agent knows how to
- 2:39repeat. And then there's memories. Those
- 2:42are everything that they've learned so
- 2:44far about you. So learning is where you
- 2:47review what they've picked up and then
- 2:49you set your quality standards. Now
- 2:51projects keep related work together.
- 2:53Library holds everything your agents
- 2:56have made ever. Things like docs to web
- 3:00pages to images. And then marketplace is
- 3:03where you can grab ready-made stuff
- 3:05instead of building it from scratch.
- 3:07Now, that's basically the whole map. I
- 3:08hope I went through it fast enough. Now
- 3:10that we have that out of the way, let's
- 3:13build.
- 3:17So, I'm creating a new agent and I'm
- 3:20going to be naming it, let's name it
- 3:22mini Zeno. Okay, it's basically a mini
- 3:24version of me. So, for the model, I'm
- 3:27selecting GBT6 A. Now, this is the most
- 3:30important step. And to be honest with
- 3:31you, it's the it's the one that most
- 3:32people would rush. That is the job
- 3:34description. So, I'm not writing. Be my
- 3:37assistant. I'm actually going to be very
- 3:40specific. So, manage my inbox, prep my
- 3:42calendar, give me a briefing every
- 3:45morning. Then, next is the tools that
- 3:47Mini Zeno needs to actually do the job.
- 3:49So, Gmail, so that it can read and it
- 3:51can draft emails. Without it, it can
- 3:54tell me about my day, but it can't
- 3:55necessarily touch my inbox. Then we're
- 3:58going to give it calendar so that it
- 3:59knows what my days look like. Then we
- 4:01are going to give it Slack so that I can
- 4:03talk to it from anywhere and not just
- 4:05inside the app. And these connections
- 4:08are per agent. They are not global. So
- 4:10Mini Zeno gets my inbox. Future agents
- 4:14will only get what they actually need.
- 4:16So that's a real security win over here.
- 4:20Now, if a tool that you use isn't
- 4:22listed, then all you can do is you can
- 4:24just plug in a custom MCP connection and
- 4:27cover pretty much anything else. So,
- 4:30there we have that. Now, the setup is
- 4:32now done. But the question is, can it
- 4:35actually do anything useful?
- 4:41Well, we're going to find out right now.
- 4:43So, let's test it out. We're going to
- 4:44say sort my last 24 hours of email by
- 4:47urgency and draft applies to the
- 4:50important ones. Okay. Now before I hit
- 4:52enter, I'm switching over to plan mode.
- 4:55So this is a non-negotiable when you are
- 4:57starting out. So Mini Zeno shows me it's
- 5:00plan first and then all I have to do is
- 5:03I have to approve it before anything
- 5:05actually happens. And here's the result
- 5:07that we get. It's sorted everything by
- 5:10what's actually important and then
- 5:13adopted replies to the ones that need me
- 5:16the most. Now, nothing has been sent. I
- 5:19just review and then I check over it and
- 5:22then I approve. Now, that's the easy
- 5:25part. The real unlock is when it starts
- 5:28to remember things.
- 5:34Now, this is where it goes from useful
- 5:37to really, really valuable. I'm using
- 5:40Teach Me Your World to tell Mini Zeno
- 5:42about me. So, my role, my tone, and all
- 5:46the things that I really care about.
- 5:48Now, think of it kind of like briefing a
- 5:51new hire on day one. Okay? It's pretty
- 5:53simple. Now, I want you to watch this.
- 5:56One of those drafts says yes to a cold
- 5:59outreach email. Now, I don't want that.
- 6:01So what I do is I tell it always the
- 6:04client called cold outreach going
- 6:06forward. Now one correction and it's
- 6:09saved as a memory. So tomorrow, next
- 6:11week, next month it just knows how I
- 6:14want things. And that means that you
- 6:16never actually have to train it twice.
- 6:22Now let's make this morning routine
- 6:26reusable. I'm saving the whole workflow.
- 6:29checking email, checking the calendar,
- 6:30pulling my tasks as a skill and I'm
- 6:33going to be calling it morning brief.
- 6:35Now, there's no zip files, no the
- 6:36explaining the same instructions every
- 6:38day. Once it's a skill, then Mini Zeno
- 6:41knows how to do it the same way every
- 6:44single time. And also, I'm going to be
- 6:46adding a rubric. So, now that's my
- 6:48quality standard for a good brief. So,
- 6:51short top three priorities first. Every
- 6:54meeting is listed. Every time the brief
- 6:56runs, it gets scored against the
- 6:58standard. So I can actually see if it's
- 7:00getting better. Now the skills template
- 7:02and the rubric, all of it is inside our
- 7:04free school community. Okay? So that
- 7:06means that you don't have to start from
- 7:07a blank page.
- 7:13Let's make it run without [music] me.
- 7:15I'm scheduling morning brief for 6:00
- 7:18a.m. every day. Then I'm adding a Slack
- 7:20triger. So now I can act mention mini
- 7:24Zeno from my phone, my laptop, and
- 7:26anywhere. and then it will respond to me
- 7:28like a teammate. I'm turning on live
- 7:31mode instead of waiting to be asked.
- 7:33Now, Mini Zeno keeps checking on its
- 7:36own. So, if an urgent email lands at,
- 7:39let's say, 2:00 p.m., it then flags it
- 7:41without asking me. Now, here's what I
- 7:44wake up to. That done, it cost me a few
- 7:47dollars. A human assistant doing this
- 7:50would would cost you thousands of
- 7:52dollars per month. That's a full
- 7:54personal assistant agent. memory, tools,
- 7:57a schedule, a real work done while I am
- 8:00fast asleep. But it's still one agent
- 8:03doing everything. Let's give it a team.
- 8:10And I mean, we all know that one agent
- 8:13is good, but a team is even better. Now,
- 8:15you already know how to build one. A
- 8:17clear description, the right tools, and
- 8:19memory. Now, we just do it a few more
- 8:21times. I'm not walking through every
- 8:23click again, so just watch how fast this
- 8:26goes. Now, first is scout, then
- 8:29research. Now, her only job is to look
- 8:33into every external person on today's
- 8:36calendar. Then next is scribe and
- 8:39writing. Now, he turns research into
- 8:42meeting prep dossier and drafts
- 8:45follow-up emails and checker review. It
- 8:47verifies the facts and it makes sure
- 8:49that everything sounds like me. Now
- 8:52remember the cost point these are
- 8:55focused repetitive jobs. So scout scribe
- 8:57and checker will run on cheaper models.
- 9:00Only Mini Zeno runs on ADA because Mini
- 9:04Zenho makes the judgment calls. Okay.
- 9:10Now here's where it comes together. Up
- 9:14to now Mini Zeno has been doing all the
- 9:17work itself. Okay. Now I'm actually
- 9:19going to I'm going to be promoting it.
- 9:22It's not just my assistant anymore. It's
- 9:24the one that manages the other three.
- 9:26Now first I'm going to add scout then
- 9:29scribe and then checker to its allowed
- 9:32agent. Now for each one I decide does it
- 9:35need my approval to run or can it just
- 9:39[music] go. Then I update Mini Zeno's
- 9:41job description with the delegation
- 9:43rule. So, every day at 6:00 a.m., check
- 9:46today's calendar. Send every external
- 9:48attendee to Scout for research. Take
- 9:50Scout's findings and brief Scribe to
- 9:53write the dossier and the follow-ups.
- 9:56Then send everything to Checker for
- 9:58review before it even reaches me. Now,
- 10:01notice the setup. The specialists don't
- 10:03pass work to each other. Everything goes
- 10:06through Mini Zeno. It's it's it's the
- 10:08hub. They are the spokes. That's how a
- 10:13deal manager works. So, it hands out the
- 10:16work. It checks what comes back and then
- 10:18it decides what's next, which means that
- 10:21I'm not the manager anymore. And right
- 10:23now, Mini Zenho is the manager.
- 10:29Now, let's watch it. Okay. The 6 a.m.
- 10:32schedule fires. Mini Zeno checks the
- 10:35calendar and starts delegating. Scout
- 10:37researches each attendee in her own
- 10:39thread and [music] then reports back.
- 10:41Mini Zeno reads what Scout found and
- 10:44then briefs Scribe. Scribe writes the
- 10:46dossier and drafts the follow-ups and
- 10:49then it reports back. Then everything
- 10:51goes to Checker for a final pass on
- 10:54facts and tone. And when I open my
- 10:57laptop, morning brief, meeting dossier,
- 11:00follow-up drafts, all of it is right
- 11:03there making my life so much easier. And
- 11:05I didn't touch a single agent this
- 11:07morning.
- 11:12Now, the part that most people don't
- 11:14even know is possible about this. I'm
- 11:17moving all four agents into a shared
- 11:19workspace. Now, I'm doing so so that my
- 11:21team can use the same agents with the
- 11:23same skills. And for memory, I choose
- 11:26what my whole team should know, like our
- 11:28brand voice, how we handle cold
- 11:31outreach, and promote it to the
- 11:34workspace. So when a teammate mentions
- 11:36scout in Slack tomorrow, scout isn't
- 11:39starting from zero. It already knows how
- 11:41we work, their conversations stay their
- 11:43own, but the knowledge is shared, which
- 11:46is so important. I also can't tell who
- 11:48can change things. Members use the
- 11:51agents, editors, and owners can edit
- 11:53them. So 10 people can rely on an agent
- 11:56without 10 people being able [music] to
- 11:58break it. Okay. And in command center, I
- 12:01see every run what it costs and how well
- 12:04each agent is scoring all in one place.
- 12:07Now, I mean, just take a step back and
- 12:09think about this. A new hire joins the
- 12:11team and on day one, they are working
- 12:13with agents that already know how your
- 12:16business runs.
- 12:21So, here's the full picture. One trigger
- 12:24at 6:00 a.m. Four agents. Mini Zeno
- 12:27running on AA managing three specialists
- 12:30on cheaper models. Research done,
- 12:33written up, fact checked, and waiting
- 12:36for me before I wake up. And the whole
- 12:40team can use every piece of it. Now, I
- 12:44don't know about you, but that sounds
- 12:45insane to me. Now, everything that you
- 12:48saw in today's video, every system
- 12:51prompt, the skill templates, the rubric,
- 12:54the exact mini Zenho delegation setup,
- 12:57all of it is for free and is inside our
- 12:59free school community. And also, you get
- 13:02a $100 bonus if you sign up for a paid
- 13:05plan on Hyper Asian. The link to that is
- 13:07down in the description below. And if
- 13:09this helped you, then what I want you to
- 13:11do right now is hit that like button.
- 13:12And also hit subscribe and turn on those
- 13:14bell notifications cuz we drop tutorials
- 13:16like this every single week. And also if
- 13:19you need another of these to turn on
- 13:20those bell notifications in the next
- 13:22video. I'm taking this even further. Not
- 13:25four agents, a whole department of them.
- 13:28So, I'll see you
About this transcript
This page contains the full transcript of How I Built a Personal Assistant Using ChatGPT-6 Astra (AI Agents) by Zinho Automates, generated from the public captions YouTube serves with the video. The transcript has 2,168 words across 306 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.