AI로만 운영되는 1인 기업 만들기 — 프롬프트 전문 (CONNECT AI LAB · AI 멘토 제이)

Starred Gists，GitHub Gist，即时分享代码、笔记与代码片段。关注

翻译

AI로만 운영되는 1인 기업 만들기 — 프롬프트 전문 (무료)

영상: 「JEV 같은 AI, 내 컴퓨터에서 0원으로 돌립니다 | AI 1인 기업 만들기 (클로드 vs GPT vs 제미나이)」 만든 사람: AI 멘토 제이 · CONNECT AI LAB (https://www.youtube.com/@CONNECT-AI-LAB)

준비 (운영비 0원)

올라마 0.35 이상 설치: https://ollama.com/download
터미널: ollama pull tev1:0.8b
확인: ollama list 에 tev1:0.8b 가 보이면 성공

사용법 3단계

아래 1_AI1인기업_템플릿_프롬프트.txt 전체를 복사해서 제미나이·챗GPT·클로드에 붙여넣기
Shift+Enter로 줄을 바꾸고, 이어서 이 문장을 넣고 전송
위의 프롬프트를 변경하지 말고, 중간중간 내용만 [내 회사 / 서비스 설명]에 맞게 변경해서 다시 작성해줘
나온 프롬프트를 클로드 코드 / 코덱스 / 안티그래비티에 넣으면 내 회사용 AI 1인 기업 완성

영상 속 예시(글로벌 AI 사주 서비스)로 바꾼 결과는 2_예시_글로벌AI사주서비스.txt 에 있어요.

참고

Nimble·Tev1은 JEV 자체가 아니라 JEV처럼 확률로 판단하는 오픈 모델입니다.
막히면 무료로 질문하세요: https://www.aicitybuilders.com/ask
제가 직접 만든 판단 AI 두뇌, Connect AI C1: https://www.ezerai.xyz/models#company-decisions

	Build "My AI-Run One-Person Company" — a one-person business where I am the only human (the CEO) and every employee is an AI.

	[OUR COMPANY] ← fill in
	Business: (e.g. A one-person AI education company on YouTube. Free videos + paid online courses.)
	What comes in: (e.g. course inquiry emails, payment/refund questions, sponsorship offers, YouTube comment questions and error reports)

	Visualize the company as a cute top-down pixel-art office where adorable AI employee characters do the work. Every decision is made by a LOCAL decision AI running on my own computer. It should make viewers say "wow" — charming, lively, polished.
	[1. DELIVERABLE]
	- ONE single HTML file. Double-click to open in a browser. No install, no build step.
	- No external image files: draw all pixel characters and the office in code (canvas, etc.).
	- All on-screen UI text must be in KOREAN.
	[2. QUESTIONS & ACTIONS — YOU design the first version]
	- For the company above, design what an expert would check (questions) and what to do with each answer (actions).
	- Structure it as a chain: ask a question → based on the answer, take an action OR ask the next question. · choice: each option → action or next question · noul (yes/no): yes / no → action or next question (yes threshold default 70%) · score: each level → action or next question
	- At any step, if confidence is below the threshold (default 0.5) → send to 👑 CEO (me).
	- Each action has an owner character (department) and a task (draft a reply, assign to someone, ask the CEO, hold).
	- Briefly explain inside the app WHY you designed it this way.
	- Let me edit it later: show the question flow as a tree diagram; add/edit/delete questions, options, actions. Adding an action automatically creates a new desk + character.
	- Save settings in the browser; export/import as JSON.
	[3. COMPANY SCREEN — VISUALS]
	- Cute top-down pixel office: one desk + character per action/department, plus the 👑 CEO desk.
	- 📬 Mailbox: messages arrive one by one as letter icons → the letter travels through each question's character, with a speech bubble above each showing that question's answer.
	- The final action's character takes the letter and works on it at their desk (show reply drafts in a speech bubble).
	- Low confidence → the letter goes to the 👑 CEO desk with a "?" bubble. When I pick the action, it goes there.
	- Walking / idle / happy animations; sparkle effect when a task completes.
	- Big counters at the top: tasks handled · avg decision time (s) · cost ₩0.
	[4. MAKE THE DECISION VISIBLE — SIDE PANEL]
	- For each question: animated probability bars per option + confidence gauge (green → red).
	- Highlight the path the current letter took through the tree.
	- Decision time (ms).
	[5. SENDING MESSAGES]
	- Type a message manually and send it.
	- 10 example messages for the company above (Korean, realistic, include 2–3 ambiguous ones).
	- "Auto-run" button: keeps feeding examples every few seconds.
	[6. DECISION LOG]
	- List: message summary, each question's answer + probabilities, confidence, action taken, time.
	- 👍 correct / 👎 wrong on each row (if wrong, pick the correct action) → accuracy % stats.
	- Export as JSON (I'll use this data to retrain later).
	[7. LOCAL DECISION AI — USE EXACTLY THIS] This is a new feature in Ollama 0.35+. Follow this spec exactly.
	- Endpoint: POST http://localhost:11434/v1/systemone (Content-Type: application/json)
	- Model dropdown: tev1:0.8b (default) / tev1 / nimble
	- Request: { "model": "tev1:0.8b", "state": { "sender": "...", "subject": "...", "body": "..." }, "questions": { "team": { "type": "choice", "instructions": "Which team should handle this message?", "criteria": { "billing": "Payments and refunds", "technical": "Bugs and login problems", "other": "None of the above" } }, "refund": { "type": "noul", "instructions": "Does the sender explicitly ask for a refund?" }, "urgency": { "type": "score", "instructions": "How urgent is this?", "criteria": ["Routine", "Soon", "Urgent"] } } }
	- Response: { "model": "tev1:0.8b", "answers": { "team": { "type": "choice", "choice": "billing", "probabilities": { "billing": 0.985, "technical": 0.012, "other": 0.003 }, "confidence": 0.922 }, "refund": { "type": "noul", "noul": 0.997 }, "urgency": { "type": "score", "score": 0.815, "legend": {"0": "Routine", "1": "Soon", "2": "Urgent"}, "probabilities": {"0": 0.378, "1": 0.429, "2": 0.193}, "confidence": 0.046 } }, "usage": { "input_tokens": 841, "output_tokens": 4 } }
	- Limits: 1–64 questions, 2–26 options each, request body ≤ 64 KB. score = probability-weighted average of zero-based levels.
	- The model works best with ENGLISH instructions and criteria. Show Korean in the UI, but send English instructions/criteria to the model (each question has a Korean label + English text field).
	- If the connection fails, show a friendly Korean guide: is Ollama running, version ≥ 0.35, ollama pull tev1:0.8b, and OLLAMA_ORIGINS if the browser is blocked.
	- NEVER fake results. If you add a demo mode for when Ollama isn't connected, show a big "체험 모드 · 실제 AI 아님" badge.
	[8. DONE WHEN]
	- One HTML file, runs on double-click
	- Question/action flow designed for the company above, with your design rationale
	- Tree view + editing; adding an action auto-creates a character
	- Real local decisions via Ollama (no fake data)
	- Characters move between desks with per-question speech bubbles
	- Probability bars + confidence gauge + highlighted path
	- Low confidence → CEO, I decide
	- Decision log + correct/wrong + JSON export When finished, tell me briefly in Korean how to run it.

	Build "My AI-Run One-Person Company" — a one-person business where I am the only human (the CEO) and every employee is an AI. Business: A one-person company running a global AI fortune-telling (Saju/사주팔자) website. It uses AI to analyze Saju globally and charges users for the service. What comes in: (e.g. birth time/date correction requests, payment/refund questions for Saju readings, translation/language errors on the global site, inquiries about compatibility/yearly fortunes, and bug reports) Visualize the company as a cute top-down pixel-art office where adorable AI employee characters do the work. Every decision is made by a LOCAL decision AI running on my own computer. It should make viewers say "wow" — charming, lively, polished.
	[1. DELIVERABLE]
	- ONE single HTML file. Double-click to open in a browser. No install, no build step.
	- No external image files: draw all pixel characters and the office in code (canvas, etc.).
	- All on-screen UI text must be in KOREAN.
	[2. QUESTIONS & ACTIONS — YOU design the first version]
	- For the global AI Saju company above, design what an expert would check (questions) and what to do with each answer (actions).
	- Structure it as a chain: ask a question → based on the answer, take an action OR ask the next question. · choice: each option → action or next question · noul (yes/no): yes / no → action or next question (yes threshold default 70%) · score: each level → action or next question
	- At any step, if confidence is below the threshold (default 0.5) → send to 👑 CEO (me).
	- Each action has an owner character (department) and a task (draft a reply, assign to someone, ask the CEO, hold).
	- Briefly explain inside the app WHY you designed it this way.
	- Let me edit it later: show the question flow as a tree diagram; add/edit/delete questions, options, actions. Adding an action automatically creates a new desk + character.
	- Save settings in the browser; export/import as JSON.
	[3. COMPANY SCREEN — VISUALS]
	- Cute top-down pixel office: one desk + character per action/department, plus the 👑 CEO desk.
	- 📬 Mailbox: messages arrive one by one as letter icons → the letter travels through each question's character, with a speech bubble above each showing that question's answer.
	- The final action's character takes the letter and works on it at their desk (show reply drafts in a speech bubble).
	- Low confidence → the letter goes to the 👑 CEO desk with a "?" bubble. When I pick the action, it goes there.
	- Walking / idle / happy animations; sparkle effect when a task completes.
	- Big counters at the top: tasks handled · avg decision time (s) · cost ₩0.
	[4. MAKE THE DECISION VISIBLE — SIDE PANEL]
	- For each question: animated probability bars per option + confidence gauge (green → red).
	- Highlight the path the current letter took through the tree.
	- Decision time (ms).
	[5. SENDING MESSAGES]
	- Type a message manually and send it.
	- 10 example messages for the global Saju company above (Korean, realistic, include 2–3 ambiguous ones like "The translation for my wealth luck is weird and I want a refund").
	- "Auto-run" button: keeps feeding examples every few seconds.
	[6. DECISION LOG]
	- List: message summary, each question's answer + probabilities, confidence, action taken, time.
	- 👍 correct / 👎 wrong on each row (if wrong, pick the correct action) → accuracy % stats.
	- Export as JSON (I'll use this data to retrain later).
	[7. LOCAL DECISION AI — USE EXACTLY THIS] This is a new feature in Ollama 0.35+. Follow this spec exactly.
	- Endpoint: POST http://localhost:11434/v1/systemone (Content-Type: application/json)
	- Model dropdown: tev1:0.8b (default) / tev1 / nimble
	- Request: { "model": "tev1:0.8b", "state": { "sender": "...", "subject": "...", "body": "..." }, "questions": { "team": { "type": "choice", "instructions": "Which team should handle this message?", "criteria": { "billing": "Payments and refunds for Saju readings", "technical": "Bugs, translation errors, and login problems", "consulting": "Questions about Saju, birth time corrections, or destiny interpretations" } }, "refund": { "type": "noul", "instructions": "Does the sender explicitly ask for a refund?" }, "urgency": { "type": "score", "instructions": "How urgent is this?", "criteria": ["Routine", "Soon", "Urgent"] } } }
	- Response: { "model": "tev1:0.8b", "answers": { "team": { "type": "choice", "choice": "billing", "probabilities": { "billing": 0.985, "technical": 0.012, "consulting": 0.003 }, "confidence": 0.922 }, "refund": { "type": "noul", "noul": 0.997 }, "urgency": { "type": "score", "score": 0.815, "legend": {"0": "Routine", "1": "Soon", "2": "Urgent"}, "probabilities": {"0": 0.378, "1": 0.429, "2": 0.193}, "confidence": 0.046 } }, "usage": { "input_tokens": 841, "output_tokens": 4 } }
	- Limits: 1–64 questions, 2–26 options each, request body ≤ 64 KB. score = probability-weighted average of zero-based levels.
	- The model works best with ENGLISH instructions and criteria. Show Korean in the UI, but send English instructions/criteria to the model (each question has a Korean label + English text field).
	- If the connection fails, show a friendly Korean guide: is Ollama running, version ≥ 0.35, ollama pull tev1:0.8b, and OLLAMA_ORIGINS if the browser is blocked.
	- NEVER fake results. If you add a demo mode for when Ollama isn't connected, show a big "체험 모드 · 실제 AI 아님" badge.
	[8. DONE WHEN]
	- One HTML file, runs on double-click
	- Question/action flow designed for the global AI Saju company above, with your design rationale
	- Tree view + editing; adding an action auto-creates a character
	- Real local decisions via Ollama (no fake data)
	- Characters move between desks with per-question speech bubbles
	- Probability bars + confidence gauge + highlighted path
	- Low confidence → CEO, I decide
	- Decision log + correct/wrong + JSON export When finished, tell me briefly in Korean how to run it.

	[② 내 회사에 맞게 바꾸는 프롬프트]
	①(프롬프트_1인기업_템플릿.txt)을 AI(제미나이/챗GPT/클로드 등)에 붙여넣고, Shift+Enter로 줄을 바꾼 뒤 아래 문장을 이어서 넣고 전송.

	영상에서 쓴 그대로 (사주 예시):
	위의 프롬프트를 변경하지 말고, 중간중간 내용만 사주팔자 서비스를 운영하는 웹사이트 글로벌 기반으로 변경해서 다시 작성해줘

	내 회사용:
	위의 프롬프트를 변경하지 말고, 중간중간 내용만 [내 회사 / 서비스 설명]에 맞게 변경해서 다시 작성해줘

	→ 나온 결과 프롬프트를 클로드 코드 / 코덱스 / 안티그래비티에 넣어 1인 기업 만들기

添加评论

点踩

分享查看原文

参与讨论