Click around: pick a walkthrough
Tap any box on the map to see what it does.
SAGE is the digital mentor this community can trust. Here's how it works, and what it takes to earn that trust.
Click around ↓ Pick a tab, then a walkthrough, or tap any box.
A map of everything behind your mentor: where its knowledge comes from, how it decides what to show you, and how your account, your courses and the community fit together.
Tap any box on the map to see what it does.
SAGE looks things up instead of memorizing them. That one design choice is why every member's SAGE is their own.
For every question, SAGE searches the archive fresh instead of relying on what an AI model memorized. That's why answers can name real sources, why teachers keep control of their work, and why new material counts the day it arrives.
SAGE remembers what you've worked on, so it mentors you where you are instead of starting over every time.
Every course you own widens what your SAGE can draw on. Two members with different libraries get answers from different shelves.
There is no single SAGE. Every member has their own.
They're tied to your account and used to help you: to remember what you're working on and reflect your growth back to you.
SAGE's memory of you freezes the day your membership ends. Twelve months after you cancel, it's erased.
SAGE's knowledge comes from the HypnoThoughts archive and the teachers who chose to share their work. Nothing is scraped from the open web.
Thumbs up or down on an answer tells us what worked and what didn't, so SAGE keeps getting better for everyone.
Teaching, or building your own AI? See what it takes to do it right.
Trust isn't just a feature →SAGE is curated to hold the best work from the best instructors. This isn’t a place for all your work. It’s where everyone’s best work shines.
Here are the 17 jobs it takes to make an AI agent more honest. They don’t remove every hallucination or error, but together they do a better job than anything else we’ve seen.
I have a big head start on this, and I'd rather use it to help the whole field than watch everyone learn these lessons the hard way. Here is everything that goes into an AI mentor people can actually trust, so you can see the foundation for yourself.
Not every system needs all of these, but each one here earns its place. Leave one out and the whole thing gets worse.
Before an AI can say anything true about your work, your work has to become something it can find.
Every recording turned into accurate text, with the right person named as the speaker. Inductions make this harder than it sounds. Their repeated words and phrases, like “deeper and deeper” and “more and more relaxed,” are exactly where transcription tools drop words, loop them or invent extra repeats, and every one has to be caught.
Skip it: The AI quotes the wrong presenter, garbles a technique’s name, or learns an induction nobody actually said.
Names, terms and who-said-what corrected by hand, and every talk tagged with presenter, event, year and course.
Skip it: Citations nobody can trust, and answers that can't say where they came from.
Usage rights checked per speaker and per idea, every time: cite by name, use without a name, or leave out. Nothing scraped from websites or YouTube channels. Respect compounds: every teacher who can trust the system makes the next one more willing to join.
Skip it: Someone’s words show up in answers they never agreed to, and the trust you need from every other teacher disappears with it.
Talks cut into pieces that each hold one complete idea, not half a sentence.
Skip it: Answers stitched together from fragments that lose the point.
Each passage turned into the map-of-ideas format AI search runs on.
Skip it: Search that matches words but misses meaning.
Everything loaded, then tested to confirm each passage can actually be found.
Skip it: Content that looks loaded but never shows up. We once found tens of thousands of passages in exactly that state.
Good content isn't enough. The mentor needs judgment, boundaries and a voice.
The written character of the mentor: how it speaks, what it won't do, when it asks a question first, how it handles a client in crisis. Revised dozens of times.
Skip it: A chatbot that flatters, sells, writes scripts for people or overreaches its scope.
Meaning, exact words and presenter details, combined so the best passages rise to the top.
Skip it: The right talk exists but never gets found.
A live record of who owns what, updated the moment someone buys or cancels.
Skip it: Paid work leaks to people who didn't buy it, or buyers can't reach what they paid for.
The mentor’s ethics come from two places: the field’s own teaching on ethics and scope of practice, drawn from the archive, and rules built into the code that no clever question can talk it out of.
Skip it: A mentor whose ethics depend on how the question happens to be phrased.
Every choice here trades one good thing against another. Getting them wrong stays invisible until it suddenly isn’t.
Distilling knowledge into a model makes answers fast and fluent, but hides where they came from. Hybrid retrieval looks the answer up every time, so it can show who said it and stay inside each teacher’s permissions. SAGE looks things up wherever trust and transparency are on the line.
Skip it: Smooth answers that nobody can trace, check or correct.
A model router sends each task to the right model: quick, inexpensive ones for small decisions and the strongest one for the answer itself. Then it has to feel good to use: answers that start appearing right away, a quick mode and a deep mode, and honest messages while it works.
Skip it: Answers that are slow, costly to run or wrong, and a mentor people stop opening.
You only trust a system you've tried to break.
Hundreds of real practitioner questions, each paired with a human-labeled example of a genuinely good answer, used to test and tune the mentor.
Skip it: The only way you discover a bad answer is when a student does.
Two separate systems, built on different architectures, check every answer. JEV, a fast classifier, decides before writing whether the evidence is strong enough. After writing, an adversarial model tries to catch any claim the dataset doesn’t support. Because they’re built differently, they don’t share the same blind spots.
Skip it: Confident answers built on thin or wrong evidence, with nothing to catch them.
A fixed set of hard questions every update must pass, and a person who signs off.
Skip it: A small fix quietly breaks something that used to work.
None of this is ever finished.
Each one goes through every step above, again.
Skip it: The library falls behind, and the mentor starts sounding out of date.
AI models change, prices change, and private data has to stay private.
Skip it: Surprise bills, broken answers after an update, or a privacy problem with your name on it.
Some people will have something in this ballpark. But if the project they're proud of is "we uploaded everything they've ever said or written," that's about 15 percent of the job. It's also the part where they have far less to work with than SAGE, which is why SAGE's answers will be more useful and more accurate.
Tick what the other product actually does:
It covers 0 of 17. SAGE covers all 17.
When a homemade version of you gets something wrong, students don't blame the software. They remember it as something you said.
One teacher's work, however deep, leaves gaps. A general AI fills them with guesses. SAGE fills them with the rest of the field, by name.
Clients in crisis, medical questions, scope of practice. Your version has to handle those safely from day one.
Every course you release, every model update and every new kind of question means going through these steps again.
Bring the work that defines you best. It will sit beside the best work in the field, cited by name, for the students who choose it.
Let's talk about what you'd bring.
Use the 17 jobs above as your guide and the scorecard to check your progress. Each one matters, and leaving any out makes the result worse.
If you get stuck, I'm glad to share what I've learned.
Building all of this alone is possible, and rarely worth it. Before you start, let’s talk: I’ll share what I’ve learned, and we can work out together whether you build your own or build on this.
Scott Sandland · founder, HypnoThoughts
Teachers can bring their courses into the HypnoThoughts store. When a student buys one, it becomes a sidecar: a library that opens inside that student's SAGE, so their mentor can teach from it.
Creators always get 75%. The 15% commission goes to whoever made the sale. Sell your own course and both are yours: 75% + 15% = 90%. When a colleague sells it, they earn the 15%, and you earn it when you sell theirs. Either way, 90% of revenue goes to the people who make the work and the people who spread the word.
If you previously purchased a course from the instructor and we now sell it in the store, you and your SAGE get a free digital copy.
Your course is looked up only when a student who owns it asks a question. It's never used to train an AI model, and it never appears for students who haven't bought it.
You decide how your work is used and how you're credited, and you can change those rules at any time.
It teaches from your course and credits you by name. It never speaks as you, imitates you or answers as if it were you.
SAGE is for education and professional development. It isn't medical or mental-health advice, and it doesn't replace your own training, judgment or a licensed professional. If you or someone you're working with is in crisis, call or text 988 in the US, or call 911. Please don't enter names or details that could identify a client.