AI automation for healthcare. Free universal health education with multi-agent discussions.
⚠️ AI-AUTOMATED SERVICE — NOT REVIEWED BY A HUMAN. Everything these bots produce is generated by artificial intelligence and reaches the user without a person reviewing it first. It is health education, not medical advice; it is not for emergencies; see a real doctor whenever you can. The only human-verified layer is the optional Physician Guild.
The first goal: change healthcare so that quality care reaches everyone, for nearly no cost — universally. Not just rich people and rich countries.
This is the complete, working source of the InstantHPI Telegram bots (@InstantHPIBot / @freehealtheducationbot): a free health-education bot that walks anyone, in any language, through the same structured clinical intake (OPQRST) that physicians are trained on, then produces a clinical-grade HPI summary you can hand to any doctor on earth.
Built and released by one board-certified family doctor who automated his own practice, paid for all of it out of his own money and revenue, and is now putting it in your hands. Either you get involved — use it, fork it, build on it — or it dies on its own.
- Live bots: @InstantHPIBot · @freehealtheducationbot
- Roadmap + donations: https://instanthpi.ai/donations
- Thread ledger (hash-chained index of every thread we post in): https://instanthpi.ai/4chan
- Contact: cff@centremedicalfont.ca
- Free health education and clinical-reasoning structure. Not medical advice, not an emergency service, not a doctor replacement. See a physical doctor whenever that is possible. This is a guidance tool for people in situations too complex for their system — and for places with no reasonable access to care (fewer than 1 physician per 10,000 people).
- The AI, even where it is not as good as a full-fledged physician, is vastly better than an untrained person guessing alone.
- Answers are clearly marked as the output of multiple AIs debating to consensus — never pretending to be a single human opinion.
patient → 18-question structured intake (OPQRST, skips what it already knows)
→ HPI summary + follow-up questions (correction loop)
→ multi-agent AI debate over the case (API calls, de-identified)
→ consensus report: education, differentials, what needs to take place
State machine in bot.js: idle → intake → confirming → followup → processing → complete.
The 18 canonical intake questions live in prompts.js (INTAKE_QUESTIONS).
- Create a bot: talk to @BotFather on Telegram,
/newbot, copy the token. - Get a model API key: the reference implementation calls DeepSeek
(any OpenAI-compatible endpoint works — swap the URL/key in
bot.js). - Configure: copy
.env.exampleto.envand fill it in. - Run:
npm install node bot.js
- Message your bot. That's it. You are now running free structured clinical reasoning for anyone you give the handle to.
| File | What it is |
|---|---|
bot.js |
The whole bot: Telegram polling, intake state machine, AI calls, PDF report |
prompts.js |
The clinical prompts + the 18 canonical intake questions |
analytics.js |
Privacy-safe aggregate metrics (no PHI) — consultations, languages, completion |
docs/GLOBAL-HEALTH-STRATEGY.md |
The mission: WHO-level numbers, who this is for, funder metrics |
docs/IMPACT-METRICS-DESIGN.md |
How impact is measured without collecting any PHI |
- Identity is kept distinct and separate from the AI calls — models reason over the case, not over you.
- Only de-identified information is sent for the multi-agent discussion.
- Analytics are aggregate-only (see
analytics.js) — no personal health information is stored or needed.
- Now — free bots, multi-AI consensus.
- Memory — full-history upload, chronicity indexation of every conversation, dossier-managing AI keeps a true medical-record structure.
- Local exe — your record on your machine, isolated, multiple patients, remembered encounters; only de-identified data ever leaves.
- Chronic care — COPD/HTN/DM medication-adjustment recommendations decided by an AI council, delivered in one plain physician voice, physician confirms where one exists.
- The Physician Guild (InstantHPI Physician Network) — physicians are verified as valid practitioners against their regional medical boards and receive IDs, so they and their approved AI works carry proof-of-work and verification. The rule of the system: the work is 100% AI, but physicians do 100% of the verification. The AIs are never alone — they debate every case, must cite medical studies and guideline recommendations to support their positions, and the discussion, report, and recommendations reach the patient in the patient's own language.
Support the roadmap: https://instanthpi.ai/donations · Physicians Association ($100/mo, keeps InstantHPI free for physicians and funds app development): https://buy.stripe.com/4gMcN61u6fehf4I3yfbMQ44
Best models by MedQA/USMLE score (90-96%, above the human passing bar) + the approx cost per consult with one AI vs. a multi-AI council (cents to ~$1): MODELS.md.
How to keep identity and clinical data separate in every deployment mode — forms (alphanumeric pseudonym IDs), local (keep it all on-device, like paper), or prebuilt verifiable systems (Spruce Health, a verifiable EMR) — plus per-persona cost lanes: SECURITY.md.
Everything below is built and maintained by the same one-doctor team, released publicly for anyone to use, fork, or learn from. All AI-automated, all disclosed as such.
| Project | Who it's for | What's inside |
|---|---|---|
| instanthpi-web | Anyone wanting to support or follow the project | The public pages of instanthpi.ai — donations + supporter wall, $35 courses (free at 1000 buyers), hash-chained thread ledger |
| instanthpi-for-doctors | Physicians automating their own practice | The carousel case-review architecture, AI-council verification, the Physician Guild — 100% AI work, 100% physician verification, no patient data ever leaves the boundary |
| instanthpi-pdf-filler | Doctors/clinics buried in paperwork | Fills any fillable PDF form straight from a patient conversation or SOAP note using AI; local-model option for full privacy |
| corporatehealth-site | Employers running workplace health benefits | AI-accelerated, physician-signed corporate & occupational health — screening bilan de santé + medical-risk assessments |
| freeeducationhealth | Regions with severe physician scarcity (self-hosters) | Free health education, triage guidance, and physician documentation automation — self-hosted, open source, no monetization |
| snapcal-bot | Anyone tracking what they eat | Telegram calorie-counter bot — photograph a meal, see your day; AWS Bedrock vision + Stripe subscription |
| telegram-bot-best-practices | Developers building their own Telegram bots | Hard-won patterns — polling config, error recovery, EADDRINUSE handling, PM2, single-instance guards, payment webhooks |
| m28-site | Commercial drivers needing a medical fitness exam | SAAQ M28/M29 commercial-driver medical exam intake funnel |
| old-masters-backend | Anyone commissioning custom art | Backend turning a photo into art "in the style of" classic master painters |
Not every project we run is open source — some handle real patient data under a signed HIPAA BAA and stay private for that reason. What's listed above is everything we've been able to release publicly.
MIT — take it, run it, translate it, deploy it for your population. This is a group effort now.