AI in Dr Non's style — not him. Verify before you act.
01
Paste it
Draft, AI paste, photo, or PDF. It doesn't have to be good.
— NON SCORE Details
02
Pick the line
Presets set the dials. Override anything.
Intent optional
Dials register · heat · length · language
Register
How hard it lands
How long
Language out
Same as the source, unless you pick another.
Anything else optional
03
Sent to a language model (cloud free tiers or a model on this Mac). Saved forever for study — nothing private. Fine print.
Service key (optional)
Your key stays in this browser, is sent only for your job, and never enters the corpus. How-tos live under the provider list and on System.
Paste something, then check the box.
This one lands hard
It is for you, not to send. Confirm.
What's Non-Writing? —
One test: is this more meaningful than silence? Cut what fails. Gold standard: make the complicated thing simple — never the reverse. Jobs live on System. Laws below.
The canon underneath
From writers he trusts on clarity. Comprehension is the point of language.
Take it with you
Same rules the engine runs, as Markdown. Hand it to any model.
01
Ask it
A straight question. Sources when it searches.
02
How hard should it land?
Delivery, not facts.
Language answer language
Same as the question, unless you pick another.
03
This is not Dr Non — a machine imitating how he argues. Not his opinion, not his employer's, not the Thai government's.
When the question needs current or contested facts, it searches the open web and cites what it read. For Thai statutes and ฎีกา with checkable links, use Non-Laws. It can still be wrong — most dangerous when wrong confidently. Not medical, legal, or financial advice. Question and answer are saved forever for study (same corpus as rewrites). Don't paste anything private. Detail: Fine print.
Service key (optional — Gemini restores search)
Shared with The Non-Writer. Edit it there.
Type a question, then check the box.
Answers land here. Ask above → Ask.
คำตอบอยู่ฝั่งนี้ — พิมพ์คำถามด้านบน แล้วกดถาม
A
In the queue
—
in the queue
Searching, then thinking.
Service change
How Non-Laws answers you
Corpus first. Story second. Clever third — and you lose.
- Substance from iApp. Licensed Thai Legal Data — real มาตรา and ฎีกา, every citation checkable against the original.
- Told in Dr Non's voice. The pack is rewritten with Non-Writer storytelling — the kind that earns trust instead of jargon.
- An accidental splash of knowledge that doesn't stop. Answer first → why the law exists → your cleverest counter-argument, steelmanned, then dismantled → what you can cite. You didn't just get an answer. You learned a lot.
- You're not that clever after all. The appendix on the right is the receipts — verbatim lines you can cite for real, or to show off.
01
Ask the law
Your rights, why they exist, and the counter-argument you thought was clever.
Answered already — no waiting, no API needed
Five invented, generic scenarios — never a real question. Run once for real citations, then frozen, so one still loads if every engine below is down.
02
How hard should it land?
Delivery, not facts.
Language answer language
Same as the question, unless you pick another.
03
Research aid, not a lawyer. iApp supplies the มาตรา and ฎีกา; Dr Non supplies the story. Check every citation. Not licensed advice; not a substitute for a Thai ทนายความ. Saved forever for study. Fine print.
Service key (optional — Gemini restores search)
Shared with The Non-Writer. Edit it there.
Type a Thai-law question, then check the box.
The answer lands here. The appendix on the right fills with the lines you can cite.
คำตอบอยู่ฝั่งนี้ — ภาคผนวกด้านขวาคือตัวบทที่อ้างได้
L
In the queue
—
in the queue
Pulling the corpus, then writing the story.
Story in Dr Non's style — receipts in the appendix. Not licensed advice.
Service change
What the desk does before you wake up
Everyone else sells you the day. This one files it, scores it, and lets you argue back.
- The desk reads first. Feeds, Hacker News, Thai press, repos — pulled overnight by Dr Non's Digest, the sibling service on this same machine.
- Every story carries its dials. Signal is how much it matters. Hype is how loudly it was sold. Fake is how much of it will not survive a week. Numbers, not adjectives.
- You pick one. The board is a route, not a feed — most meaningful first, and nothing filed that is not more meaningful than silence.
- Then you interrogate it. Ask Non answers about that exact story with the story in hand: point first, argument, counter-argument, synthesis.
01
The board
What the desk filed, newest first. Tap a story to open it.
Reading the desk…
A few minutes, once a day, out loud
Everything else reads you the headlines. This one tells you which ones earned the airtime.
- Two feeds, not fifty. Hacker News is what builders are arguing about. The Economist is what the money is watching. Headlines and blurbs only — nothing behind a paywall is copied.
- Dr Non writes it for the ear. Not a summary read aloud — subject and verb early, one idea per sentence, and an honest "the headline is all I have" when that is the truth.
- Then it is spoken in the cloned voice. Synthesised on this machine from a recording of Dr Non. It is a clone, and the page says so every time.
- Every episode leaves a trace. Each day files what it suggests about where attention is going. Enough days and the archive shows a shape no single morning does.
01
Today's episode
Cut once a day from the morning's feeds. Old ones stay below, forever.
Looking for today's episode…
No audio today — the voice engine did not answer. The script below is the whole episode.
Read it instead
02
The archive
Every day since this line opened. Tap one to hear it again.
Ask a question. It goes and reads.
A search engine hands you ten links. This reads them, connects them, and tells you what it found.
- It plans first. Your question becomes search queries — in the languages that actually carry the answer, not just English.
- Then it reads, properly. Nine engines behind one chain: Crawl4AI drives a real browser for JavaScript pages, Scrapling survives redesigns and basic anti-bot, MarkItDown handles documents. It falls through until something works.
- It connects what it read. Entities and relationships across every page, so you see how the sources agree, disagree, and cite each other — not a list of ten blue links.
- Then Dr Non narrates it. Same laws as everything else here: most important thing first, nothing invented, sources you can open.
01
What do you want to know?
A real question, not keywords. It plans its own searches from this.
Start from specific pages optional · up to 25 URLs
One per line. Give it nothing and it finds its own way in.
02
How far should it go?
More pages means a better answer and a longer wait. These runs take minutes, not seconds.
Pages to read
How deep to follow links
Language what it reads · what you get back
Search in
It picks the languages that carry the answer. Leave this alone unless you know better.
Answer in
Same as the question, unless you pick another.
03
This reads the open web on your behalf. It honours robots.txt, refuses private and local addresses, and will not touch onion services. Pages it reads are stored on Dr Non's machine — your question and everything it fetches are kept for study.
These runs are slow on purpose: dozens of pages, minutes per job, ten per hour. Fine print.
Ask a question, then check the box.
What it found lands here. The sources and the map fill on the right.
สิ่งที่มันค้นเจอจะอยู่ฝั่งนี้ — แหล่งที่มาและแผนผังอยู่ด้านขวา
S
In the queue
Planning the search.
Written from pages it actually read. Open the sources — it can still misread one.
Service change
Say what you want to learn. Get a short class.
A topic, or a document you need to understand — turned into ten minutes of teaching, argument, and a check.
- It reads first. Paste or upload your own material and it teaches only from that. Give it just a topic and it searches the open web, and cites what it found.
- Dr Non teaches. Short lessons that show you a thing before naming it, with the one small detail that turns out to matter.
- The class argues back. A skeptic raises the strongest objection; a novice asks the question you were embarrassed to ask. Raise your own hand on any scene.
- Then it checks you — honestly. Every quiz answer must quote its source, and the machine checks the quote is really there. What you miss comes back at the end.
Pick up where you left off:
01
What do you want to learn?
Type a topic or a question in the box. Have a document? Upload a photo or PDF, or paste it below — it will teach from that and nothing else.
Teach me from my own material optional · paste text, or use Upload above
0
characters
/ 12000 max
With material here, every fact it teaches has to come from it. Under 200 characters counts as part of your request.
02
Teach me in
03
Your request, anything you paste or upload, and the course are saved on Dr Non's machine for study — not published, not shown to other users. Courses are built on free-tier cloud engines; with no material, it searches the open web.
A machine-made course can still be wrong. It shows you its sources and flags any figure it could not find in them. Fine print.
Type what you want to learn in the box above.
Service change
In the queue.
Sources it taught from
Where was this photo taken?
For the place you remember but can no longer name — read from the photo itself, then checked against the map.
- It reads the file first. If the photo still carries GPS or a date, that is a lead — but only a lead. Metadata can be wrong; the picture wins.
- Then it reads the picture. Signs and text first, then plates, which side cars drive on, buses, wires, roofs, plants, mountains — the clues that rule out the most ground come first.
- Every guess is checked. Each candidate is looked up on OpenStreetMap and against tables of driving sides and phone codes. A name the map cannot find gets no coordinates.
- You get a range, honestly rated. Country, city and street are rated separately, with an error radius, the clues for and against, and what would settle it.
01
Your photo
The original file is best — screenshots and photos sent through chat apps lose their GPS and date.
Drop a photo here, or tap to choose one
JPEG, PNG, WebP or HEIC · up to 10 MB
02
Why do you need to know?
Finding where a photo was taken can be used to find a person. So Non-See asks before it looks. Pick a reason and say it in your own words.
0 / 15+
03
Use this only in good faith. Do not use Non-See to find, follow, watch, harass or expose a person, to find where someone lives, or to locate a child. Doing so can be a crime, and it is not what this tool is for.
Photos of a home, a person or a child, from anyone who did not choose "My own photo", get city-level answers only. Your reason and the result are saved for study; the photo itself is not saved. It is sent to a cloud vision engine to be read. Answers can be wrong. Fine print.
Start with step 01: why do you need to know?
Service change
In the queue.
Was this right?
You may know more than the machine. Every answer you give here teaches Non-See for the next photo.
This photo shows a home, a person or a child, and it is not marked as your own photo — so the answer is limited to city level.
The cloud vision engines were busy, so the small model on this machine read the photo. Treat this as a guess and try again later.
How sure, level by level
Candidates — the board
What it saw
Photo metadata — a lead, not proof
What would settle it
What was actually run
Not run here: reverse image search, satellite matching, street-view matching. If the answer matters, try those yourself on the top candidate.
Method after geo-sleuth by Oldcircle (MIT). Map data © OpenStreetMap contributors (ODbL). Lookup tables from Wikipedia (CC BY-SA 4.0).
Live notes from your microphone — on this machine.
Press record, talk, and a transcript builds as you go. Ask for notes at the end.
- It listens in short pieces. Every eight seconds a piece of audio goes to Whisper, an open-source speech model running on Dr Non's machine — not a cloud service.
- The transcript trails a little behind. Expect seconds of lag, more when the machine is busy. Nothing is lost; pieces wait their turn.
- Translation is optional, and slower. A local model translates each line as it can. It is for following along, not for the record.
- Notes at the end. Summary, decisions, action items and open questions — only from what was said. Copy or download everything.
01
Before you record
Choose the language people will speak. "Detect" works, but naming it is more accurate — especially for Thai.
Speed or accuracy
02
Ask before you record. Recording people without their knowledge can be illegal, and in Thailand voices are personal data under the PDPA.
The audio is transcribed on Dr Non's machine and deleted as soon as each piece is read. The transcript and notes are kept in your browser only — the server records just that a meeting took place and how long it was, never what was said. Fine print.
0:00
Tick the box above, then press record. Your browser will ask to use the microphone.
This browser cannot record audio. Use a recent Chrome, Edge, Firefox or Safari.
Service change
Transcript
Listening… the first lines appear after about ten seconds.
Speech by Whisper (OpenAI, MIT) via mlx-whisper. Accurate uses Typhoon Whisper Turbo, a Thai fine-tune of Whisper large-v3-turbo (SCB 10X, MIT). Translation and notes use a local Ollama model. A machine transcript has errors — check names and numbers before you rely on them.