How To Run Kimi K3 Agent OS For FREE

At a glance

Julian Goldie runs a five minute setup on camera and the whole video is that setup plus the reasoning behind three settings. Go to kimi.com, open a free account, and the model dropdown lands you on K2.6 by default. Switch it to standard K3. Do not switch it to K3 Swarm, which sits right beside standard K3 in the same list and, on a free account, will drain your entire free allowance almost instantly, sometimes inside a single task. Leave the context window on standard, because extra long is locked behind the premium plan anyway. Leave thinking effort on standard and only raise it to high or max when a build genuinely needs deeper reasoning, because max makes the model think longer before it answers and that costs noticeably more tokens per response. Then you prompt it, and he prompts it with build a website for an SEO agency. On his own private benchmark, which he calls GoldyBench, he says K3 outperformed Fable 5, beat GPT 5.6, and beat Opus 4.82. He runs his standing 3D racing game benchmark on it inside Agent OS and gets a neon styled game with a working score counter, a speed display, and collectibles along the track. Two limits close the video: during peak times the free account can get paused or capped, so work off peak, and although Kimi K3 is set to go fully open source from July 2026, running it yourself needs serious hardware that a normal laptop or desktop does not have.

The presenter is a digital avatar, and he says so

The video opens with a stack of hooks. What if the best coding model right now was completely free. Is the model everyone is talking about actually weaker than a free one. Why is almost nobody talking about this. Then the introduction, verbatim and unhedged: "Hey, this is the digital avatar of Julian Goldie."

That is not a throwaway. The person on screen is a synthetic presenter, and the channel says so in the first fifteen seconds rather than letting you find out. Take it as context for everything that follows: this is a produced instructional asset from a marketing operation, not a screen recording of a person discovering something live.

His framing of his own job: he helps people learn how to actually use AI tools in their real work, not just talk about them. And the framing of Kimi K3 specifically is possessive from the start. "Inside Agent OS, Kimi K3 is one of the models we run." Agent OS is his own system, and K3 is a component he has slotted into it. Hold onto that, because the title of the video welds "Kimi K3" and "Agent OS" together, and only one of those two things is free.

He promises four deliverables, and the video does hit all four:

1. How to get K3 set up for free.
2. How to configure the settings the right way so you do not waste tokens.
3. What the difference is between the model versions you will see on screen.
4. What it can actually build once it is set up properly.

Plus a fifth: a real test comparing an older Kimi model against K3. And a warning attached to all of it. "Stay with me because the setup steps matter, and getting them wrong is the fastest way to burn through your free access."

Step one: sign up, but sign up in an incognito window

"Let's start simple. Go to kimi.com."

Then the first piece of genuine craft in the video, and it is aimed at himself as much as at you. If you are already signed in on your main account, open an incognito window, so you see exactly what a fresh free account looks like. His reason is presentational: that is the version most people watching will actually be using. But the reason it is good advice for you too is that a logged in session carries entitlements. If your main account has ever touched a premium plan, a trial, a promo, or a grandfathered limit, the interface you are looking at is not the interface a new user gets. Incognito strips the session cookie and you see the real floor.

Practically, on macOS that is Command Shift N in Chrome or Command Shift P in Firefox. On Windows, Control Shift N. Then paste the address in fresh.

When you land, the site asks you to sign in. Use a free account. He is explicit that no premium plan is needed for any of this, and nothing later in the video walks that back.

Step two: the model dropdown, and the one switch that matters

Once you are in, look at the model dropdown. It is set to K2.6 by default. That is not what you want. You want K3.

This is the entire reason the video exists. The default is not the model everyone is talking about. If you sign up, type a prompt, and judge Kimi on the answer, you have judged the previous generation model and never touched the new one. The switch is one click and almost nobody makes it, which is close to his opening question of why nobody is talking about this.

The K3 Swarm trap

"Here's where most people trip up, and it's the part that actually matters most."

Sitting right next to standard K3 in that same dropdown is an option called K3 Swarm. He is precise about why it catches people. It sounds more powerful because of the name, and it is easy to assume bigger is better.

What actually happens on a free account: you do not have enough tokens to run K3 Swarm properly, and you will burn through your entire free allowance almost instantly, sometimes within a single task.

He does not explain the architecture behind the name, so here is the reading rather than a claim he makes. A swarm mode in an agent product almost always means fanning a task out across many model instances working in parallel, each with its own context, then reconciling the results. That is why it is more capable, and it is also exactly why it is expensive: you are not paying for one reasoning pass, you are paying for N of them plus the reconciliation. On a metered free allowance, a mode whose whole design multiplies calls is the single fastest way to hit zero. His observed symptom, an entire allowance gone inside one task, is consistent with that shape, but he reports the symptom and not the mechanism, so treat the mechanism as inference.

The instruction is unambiguous: switch to standard K3. That is the one that lets you get real work done across multiple prompts without running out straight away. Notice the framing of the benefit. It is not that standard K3 is better output. It is that standard K3 gives you more turns, and on a free tier turns are the scarce resource.

Step three: context window, keep it on standard

While you are in the settings, check the context window. Keep it on standard, not extra long.

Then he immediately tells you why this is not really a decision. The extra long option is locked behind the premium plan, so on the free tier standard is the only choice available to you anyway. His stated reason for raising it at all is honest and useful: it helps to know that up front, so you are not confused when it is greyed out.

Worth spelling out what the setting is, since he does not. The context window is how much text the model can hold in view at once: your prompt, its own prior turns, any files you have pasted, and the reasoning it has produced so far. A longer window lets a single conversation carry a bigger project without forgetting the start of it. It also costs more, because every token in that window is re read on every turn. That is why extra long is the paid option and not the default. On a free account it is not a lever, it is a locked door with a label.

Step four: thinking effort, and the only dial you actually control

Three options: standard, high, and max.

Max means the model thinks longer before it answers, which uses noticeably more tokens per response. His rule: keep it on standard for simple tasks, and only bump it up when a task genuinely needs deeper reasoning, like a more complex build.

Again, the mechanism is worth naming. Reasoning effort settings control how many tokens the model is allowed to spend thinking privately before it produces its visible answer. Those thinking tokens are billed like any other. On a metered account, raising the effort dial raises your burn rate on every single response, whether or not that particular response needed the extra depth. That is why the default matters so much: standard is not the cheap compromise, it is the correct setting for the majority of prompts, and high or max are what you reach for deliberately on the one prompt out of ten that has real structure to work out.

This is the only one of the three settings that is genuinely a free tier decision. The model selector has one right answer. The context window is locked. The effort dial is yours, on every prompt, forever.

Step five: your first prompt

Once K3 is selected, the context window is standard, and thinking effort is set, you are ready to prompt.

The example he types:

build a website for an SEO agency

Five words. No role framing, no output spec, no constraints. And "it gets straight to work."

That is the demonstration and it is deliberately unimpressive, because the point of the segment is not the prompt, it is the clock. "That's the entire setup, 5 minutes and you're running one of the strongest free models available right now."

K3 or K2.6: when to reach for which

He does not treat the older model as dead weight. The split is explicit and it is the split he uses professionally.

If you want fast, instant replies for something simple, K2.6 is decent enough, since it does not need to think as hard. If you are building something that needs real structure and quality, that is when K3 takes over.

Then, the tell that this is an operational rule and not a talking point: "Inside Agent OS, this is exactly the split we use, K2.6 for quick answers, K3 for anything we're actually building."

Read that as a general pattern, because it generalizes past Kimi. In any tiered model lineup, the cheap fast model is not a worse version of the expensive one, it is a different tool for a different shape of request. Lookups, rewrites, quick classification, one line answers: cheap model. Anything with architecture in it: expensive model. Routing by task shape rather than always reaching for the strongest thing available is most of what "not wasting tokens" actually means in practice.

The benchmark claim: GoldyBench

"So, how does it actually perform? On personal testing using GoldyBench, K3 did really well."

The specific claims, as stated:

• It outperformed Fable 5.
• It beat GPT 5.6.
• It beat Opus 4.82.

His own commentary on the weight of that: "That's not a small claim, those are some of the strongest models out there right now, and a free model going toe to toe with them is genuinely worth paying attention to." And the qualitative read: it is not a stripped down option built to look good on paper, it is a model that holds up when you actually put it to work.

GoldyBench is his personal benchmark. He gives no scores, no task list, no methodology, no sample size, and no link. It is a private evaluation reported as a result. That does not make it wrong, and personal benchmarks are often more predictive of your own work than public leaderboards are, precisely because they test the tasks you actually do. But it is one person's unpublished harness reported as three wins with no numbers attached, and it should be filed as an impression rather than a measurement. If you want to check it, the honest move is the one he models later in the video: keep one build task you know well, and run it on every new model.

What K3 can actually build

Three categories, stated:

• Full games.
• Detailed 3D worlds.
• Video content, if you teach it a tool like Remotion, by combining Remotion with Kimi.

That third one is the interesting one and it deserves unpacking, because he waves at it in a single clause. Remotion is a React library for making videos programmatically: you write components, you get frames, you render an MP4. Which means video generation becomes a code generation problem, and a code generation problem is exactly what a strong coding model is for. "Teach it a tool like Remotion" means giving the model the library's API surface so it can write valid Remotion compositions rather than guessing at them. The pattern generalizes to any code as output medium: if the artifact you want can be expressed as a program, a coding model can produce the artifact.

His conclusion from the three categories is the reason K3 sits inside his system at all: "It's not just a chatbot that answers questions, it's a model built for actually producing things."

The live test: a 3D racing game, run twice

This is the strongest segment in the video because it is a controlled comparison rather than a claim.

He keeps an old benchmark he has used for a long time to compare models: a simple 3D racing game build. He runs it every time he wants to see how a new model actually holds up. Same prompt, same test, same setup inside Agent OS. The only variable he changes is the model.

He ran it previously on the older Kimi model. He ran the exact same test again on Kimi K3 inside Agent OS. Completely different result.

What the K3 run produced, itemized:

• Better structure in the generated code.
• Smoother movement.
• Way more polished overall.
• A neon styled 3D racing game.
• A working score counter.
• A speed display.
• Collectible objects scattered along the track.

"Same prompt, same test, same setup inside Agent OS, just a different model. And the jump in quality speaks for itself."

His reading of why it matters: it shows this is not a small upgrade, it is a real leap in what the model can produce when you give it a building task inside a full system like Agent OS.

Two honest notes on the test, neither of which he makes. First, the benchmark prompt is never shown on screen, so the comparison is not reproducible by a viewer. Second, "collectible objects and a speed display" are features, not quality, and a model volunteering more features on an underspecified prompt is a real and useful difference but a different one from writing better code. What he shows is a genuine before and after. What he does not show is what was asked for.

If you want to run the same shape of test yourself, the reconstructed equivalent is below. It is written from what the finished build contains, not copied from his screen, and is marked as such.

Build a playable 3D racing game that runs in a single HTML file in the browser.

Requirements
- Third person camera behind a car on a looping track.
- Keyboard steering and acceleration with a sense of weight, not instant turning.
- An on screen speed display and a score counter.
- Collectible objects scattered along the track that add to the score.
- A neon visual treatment: dark environment, emissive track edges, glow on pickups.
- No build step and no asset downloads. One file I can open directly.

Output the complete file. Do not stop to ask me questions.

Keep that file. Run it verbatim against every new model you are curious about. That is the actual transferable idea from this segment: a personal benchmark you never change is worth more than a leaderboard you did not write.

Where Agent OS comes in

Roughly halfway through, the video turns into a pitch, and it does not disguise the turn. "Now, if you're serious about setting this up properly, this is exactly what we've built inside the AI Profit Boardroom."

What is inside, as stated:

• The full Agent OS zip file, ready to install.
• A complete 30 day roadmap built specifically around it, so you are not just installing it and guessing what to do next.
• Walkthroughs of setting up models like Kimi K3 inside Agent OS.
• Live coaching calls every week, where you can share your screen and ask questions about your exact setup.
• Over 4,000 members currently inside.

The home for it is aiprofitboardroom.com, which he reads out at the very end of the video, and the description points at the AI Profit Lab community on Skool for the Agent OS and Kimi masterclass.

This is the part where the video's title does the most work. Kimi K3, the model, is free on kimi.com, exactly as promised, and nothing in the setup requires a payment. Agent OS, the wrapper he keeps saying he runs K3 inside, is a paid product. Both halves of the title are true in isolation and the sentence they make together is not. No price for the Boardroom is stated anywhere in the video.

Stretching the free tier: the three levers

Late in the video he gathers the settings advice into one place, aimed at people using K3 for real projects rather than quick tests. The three levers, and the direction to push each:

1. Stick to standard K3 instead of the swarm option.
2. Keep your context window on standard.
3. Only push your thinking effort up to high or max when the task actually calls for deeper reasoning, like a build with a lot of moving parts.

The payoff he claims: you can run more tests and more builds before you hit any kind of limit, instead of burning through everything on one oversized task.

He also widens the racing game result deliberately, because he does not want you to file K3 as a games model. The same model that built that game can just as easily structure a website, build out a landing page, or put together the Remotion video workflow. "The point isn't that K3 is only good at games, it's that the same jump in quality you saw in that test shows up across pretty much anything you ask it to build, whether that's inside Agent OS or directly through kimi.com."

That last clause is the most generous line in the video toward people who are not buying anything: directly through kimi.com is named as a legitimate path, not a lesser one.

His closing argument for bothering with the setup: once you have the settings right, you are not just experimenting anymore, you are working with a model that can genuinely carry a real project from start to finish.

Peak hours, and what a free tier actually means under load

"There's a couple more things worth knowing before you go all in on the free version."

The first is capacity. During peak times, the free account can get paused or limited on how much you can use. He is fair about it and does not dress it up as a Kimi problem: this is normal for any free AI tool that has to manage demand across a huge number of users.

The symptom to watch for is responses slowing down or your access getting capped. The fix is scheduling. Use it during off peak hours instead, since you will usually get more free tokens when fewer people are online at the same time.

He does not say which hours are off peak, so this part is a practical reading rather than his claim. Kimi comes from Moonshot AI, a Beijing based lab, and the largest share of concurrent free users on a Chinese consumer AI product will be in Chinese working hours. If you are in North America or Europe, your ordinary working day already lands substantially outside that window, which is a quiet advantage. If you are in Asia Pacific, the early morning and late evening are where the headroom is. Test it rather than trusting the theory: run the same prompt at two different times of day and watch whether you get throttled.

The wider point, stated flatly, is that free tiers on frontier models are elastic, not fixed. Your allowance is not a number in a contract, it is whatever the provider can spare when you ask. That is the real texture of the word "free" in this video's title, and it is the one limit he actually flags himself.

Open source from July 2026, and the hardware reality check

"Another big piece of news, Kimi K3 is set to go fully open source from July 2026."

He frames it correctly as a big move for developers who want to run it on their own infrastructure instead of relying on the hosted version. This video was published on 20 July 2026, so that is an imminent or just landed event at the time of recording rather than a distant roadmap item. Moonshot's previous Kimi weights have gone out through its GitHub organization and its Hugging Face org, which is where a K3 release would be expected to appear.

Then he does the thing that separates this from a hype segment. He talks you out of it.

"Before you get too excited about running it locally yourself, here's the reality check. You'd need a very strong setup, serious hardware, to run this on your own machine. On a basic laptop or desktop, the kind most people are using day to day, you simply won't be able to run the open source version yourself."

The conclusion: for most people watching, using it through kimi.com for free is still going to be the practical option for a while yet.

This is correct and it is the most useful thirty seconds in the video for anyone who has ever been told that open weights means free. Open weights means the file is available. It does not mean the file fits. A frontier scale model is tens or hundreds of gigabytes of parameters that have to be resident to serve a single token, and the hardware that holds that is a rack decision rather than a laptop decision. Open source moves the cost from a subscription line to a hardware line, and for one person the hardware line is usually the bigger one. The hosted free tier is not a compromise on the way to running it yourself, it is very likely the cheapest way you will ever run this model.

The recap, as he gives it

He closes the instructional half with a compressed checklist, and it is worth reproducing because it is the whole video in seven clauses:

1. Head to kimi.com.
2. Sign in on a free account.
3. Switch away from K2.6 to standard K3.
4. Skip the swarm option unless you know you need it.
5. Keep context standard.
6. Set your thinking effort based on the task.
7. Use K2.6 for quick replies, K3 for real builds.

Plus the scheduling note: off peak hours give you more room to work with.

The complete setup, expanded

Everything below is his sequence, with the steps he waves at spelled out. Nothing here contradicts the video; it fills in what a first time user would otherwise have to guess.

The whole thing as a checklist you can paste

Kimi K3 free setup

[ ] Open a private / incognito browser window
[ ] Go to kimi.com and create or sign in to a FREE account
[ ] Open the model dropdown (it will read K2.6)
[ ] Select: standard K3          <-- the entire point
[ ] Do NOT select: K3 Swarm      <-- empties a free allowance, sometimes in one task
[ ] Context window: standard     <-- extra long is premium only, so it is locked anyway
[ ] Thinking effort: standard    <-- raise to high/max ONLY for complex builds
[ ] Send a first prompt to confirm you are on the right model

Ongoing rules
[ ] K2.6 for quick answers, K3 for anything you are building
[ ] Effort stays on standard by default; raising it costs tokens on EVERY response
[ ] If throttled, retry off peak rather than upgrading in frustration
[ ] Keep one unchanged build prompt as your personal benchmark

Where "FREE" has edges

Everything above is what the video says. This section is the accounting, and it is worth doing carefully, because the setup advice is genuinely good and the framing around it is doing more work than the advice is.

Three things follow from that table.

The word "free" in the title covers the model and not the system. Read literally, "How To Run Kimi K3 Agent OS For FREE" promises that you can run Agent OS with Kimi K3 in it, at no cost. What the video delivers is that you can run Kimi K3 at no cost, in a browser, with no Agent OS involved. Agent OS appears in the video as the thing he runs his own tests inside, and as a zip file behind a paid community. The setup he teaches works entirely without it, which is to his credit, and the title does not say so.

Not a single number appears in the entire video. Not the free allowance, not the context window sizes, not the token multiplier for max effort, not a benchmark score, not a price for anything, not a hardware requirement. Every quantity is a direction: "noticeably more", "almost instantly", "very strong setup", "usually get more free tokens". For a video whose core promise is about managing a metered resource, that is a real gap, and it is the reason this page has diagrams and comparison tables in it rather than charts. There is nothing to plot.

The advice is still correct. That is the part a cynical read would miss. Defaulting to the older model, sitting a token incinerator one line away from the right answer, and charging for reasoning depth are all genuinely how these products are built, and a new user gets all three wrong. Switch the model, leave the effort dial alone, route cheap requests to the cheap model, and retry off peak. Those four habits will stretch a free tier on any provider, not just this one. They cost nothing to adopt and the video gives them away before the pitch starts.

The last thing to note is where the video ends up. It closes on the AI Success Lab, described as free and with 85,000 members, then on the AI Profit Boardroom, which is not free and which he says has over 4,000. That is a clean funnel and it is not hidden. You get a working setup for free in the first three minutes, and the remaining five ask you to buy the system around it. Take the first three minutes. They are worth the eight.

Key takeaways

• The default is the old model. A fresh kimi.com account opens on K2.6. If you never open the dropdown, you never touch K3, which is most of why he thinks nobody is talking about it.
• K3 Swarm sits one line from the right answer and is the trap. On a free account it drains the entire allowance almost instantly, sometimes inside a single task. Standard K3 is the pick because it buys you turns, not because it produces better output.
• The context window setting is theatre on the free tier. Extra long is premium locked. Knowing that is only useful so you are not confused by a greyed out option.
• Thinking effort is the only lever you actually control. Standard by default; high or max only for builds with real structure. Higher effort spends more tokens on every response, not just the ones that needed it.
• Route by task shape. K2.6 for quick answers, K3 for anything you are building. He says this is exactly the split they run internally, and it is the single most transferable habit in the video.
• Keep one build prompt you never change. His is a 3D racing game, run against every new model. A personal benchmark beats a leaderboard because it tests your work rather than someone else's.
• A free tier is elastic capacity, not a quota. During peak times access gets paused or capped. Retrying off peak is a real fix, and it is a limit he flags himself before anyone asks.
• Open weights are a hardware bill. K3 is set to go fully open source from July 2026, and he immediately says a normal laptop or desktop will not run it. The hosted free tier stays the practical path for almost everyone.
• Free covers the model, not the system. Kimi K3 costs nothing. Agent OS, the zip file, the 30 day roadmap and the coaching are a paid product. The setup he teaches needs none of it.
• Nothing in the video is quantified. No allowance size, no context sizes, no effort multiplier, no benchmark scores, no prices, no hardware specs. The advice is sound and directional; treat every magnitude as unstated.

Chapters

0:00 Intro, a free model beating paid AI 0:14 What you will learn: setup, settings, model test 0:48 kimi.com sign up, the free account walkthrough 1:03 The model selector, switching from K2.6 to K3 1:18 Avoid this mistake, skip K3 Swarm on the free tier 1:43 Context window, keep it standard 1:48 Thinking effort, standard versus high versus max 2:20 First prompt, building your first task 2:30 K3 versus K2.6, when to use each 2:52 Benchmark results on GoldyBench, beats GPT, Opus, Fable 3:19 What K3 can build: games, 3D worlds, video with Remotion 3:50 Live test, the 3D racing game comparison 4:29 Agent OS and the Boardroom, full setup plus roadmap 5:02 Token saving tips, stretching your free usage 6:19 Peak hours warning, free access limits 6:46 Open source news, K3 going open source July 2026 7:20 Recap, the full setup steps 7:43 Join the community, AI Success Lab and the Boardroom

Notable quotes

"Hey, this is the digital avatar of Julian Goldie." (0:14)

"Stay with me because the setup steps matter, and getting them wrong is the fastest way to burn through your free access." (0:44)

"If you're already signed in on your main account, open an incognito window so you can see exactly what a fresh free account looks like, since that's the version most people watching this will actually be using." (0:54)

"You'll notice the model drop down is set to K2.6 by default. That's not what we want. We want K3." (1:10)

"Here's where most people trip up, and it's the part that actually matters most." (1:14)

"It sounds more powerful because of the name, and it's easy to assume bigger is better. But on a free account, you don't have enough tokens to run K3 Swarm properly, and you'll burn through your entire free allowance almost instantly, sometimes within a single task." (1:22)

"Max means the model thinks longer before it answers, which uses noticeably more tokens per response." (2:04)

"That's the entire setup, 5 minutes and you're running one of the strongest free models available right now." (2:30)

"Inside Agent OS, this is exactly the split we use, K2.6 for quick answers, K3 for anything we're actually building." (2:48)

"On my own tests, it outperformed Fable 5, beat GPT 5.6, and beat Opus 4.82." (3:03)

"It's not a stripped down option built to look good on paper, it's a model that holds up when you actually put it to work." (3:18)

"It's not just a chatbot that answers questions, it's a model built for actually producing things." (3:39)

"Same prompt, same test, same setup inside Agent OS, just a different model. And the jump in quality speaks for itself." (4:13)

"Doing this means you can run more tests and more builds before you hit any kind of limit instead of burning through everything on one oversized task." (5:22)

"The point isn't that K3 is only good at games, it's that the same jump in quality you saw in that test shows up across pretty much anything you ask it to build, whether that's inside Agent OS or directly through kimi.com." (5:43)

"During peak times, the free account can get paused or limited on how much you can use, which is normal for any free AI tool that has to manage demand across a huge number of users." (6:26)

"Try using it during off peak hours instead, since you'll usually get more free tokens when fewer people are online at the same time." (6:39)

"Kimi K3 is set to go fully open source from July 2026." (6:48)

"On a basic laptop or desktop, the kind most people are using day to day, you simply won't be able to run the open source version yourself." (7:08)

Resources mentioned

• Kimi, the site the entire setup happens on. A free account reaches the model selector, where K2.6 is the default, standard K3 is the correct pick, and K3 Swarm is the one to avoid on free.
• Moonshot AI, the lab behind Kimi and the K3 model.
• Moonshot AI on GitHub and Moonshot AI on Hugging Face, where the lab has published previous Kimi releases and where open weights for K3 would be expected to land after the July 2026 open source move he mentions.
• Agent OS, Julian Goldie's own system, which he runs Kimi K3 inside. The installable zip file, the 30 day roadmap, the weekly live coaching calls and the tutorials are all inside the paid AI Profit Boardroom, which he says has over 4,000 members.
• AI Profit Lab on Skool, the description's link for the Agent OS and Kimi masterclass and the video notes.
• AI Success Lab, the free community on Skool, the free course and community he points to at the end, described in the video as 85,000 members.
• Remotion, the programmatic video library he says K3 can be taught so it generates video content by combining Remotion with Kimi.
• GoldyBench, his own private evaluation harness. Not published and not linked anywhere; the only outputs given are the three wins.
• Anthropic, maker of the Fable and Opus models he says K3 beat on his tests, and OpenAI, maker of GPT.
• Julian Goldie's free SEO strategy session and 200 plus free AI SEO prompts, the other offers listed in the video description.
• Julian Goldie SEO on YouTube, the channel.