News, updates, and insights from Hamo AI.
On 14 September 2026 the Canadian Intellectual Property Office registered HAMO as trademark TMA1,445,335 for Hamo AI Technology Ltd. — nearly two years after we filed, and five and a half months after the application was approved. Here is what was registered, how it got there, and what a registration does and does not cover.
Read more →On 11 September 2026, 25 Fields Medalists warned that the way AI companies are pushing mathematics is severely misaligned with the goals of the mathematical community. We read the statement carefully — including what it does not say — line its core up against the principle Hamo holds from AI for Psychology to Psychology for AI, and then turn the same ruler on ourselves, because Hamo is an AI company too.
Read more →The origin story of Hamo's five-letter measure — from a three-sentence brief written one February evening, through a keyword table that couldn't hear 'I want to die', to an open-weights 0.6B pulse-taker with 1,677 downloads in its first 14 days. And underneath it, an old argument settled in a new place: why symbolic type systems like MBTI can't fully describe a mind, and what a learned 3,072-dimensional representation measures instead.
Read more →Twelve days after the weights went public, hamo-score-0.6b had 999 downloads — which pleased us and worried us in about equal measure. The model is a measuring instrument, designed to sit behind a deterministic crisis gate and inside a smoothing pipeline; bare weights invite exactly the deployment it was designed against. So we open-sourced the other half: one pip install for the full gate → score → smooth → bucket pipeline, one docker compose up for a reference server, and a self-check exam — 195 synthetic scoring questions plus 10 handwritten gate cases — whose very first run caught a bug in our own crisis wordlist. The code is Apache-2.0.
Read more →The companion methodology notes to our open-weights release. How a 0.6B model learned to score psychological state from a cloud API's judgments — the three-stage teaching chain, the privacy rule that shaped every decision, a plateau that more data would not shift, and the diagnosis that broke it. A model three times larger gained one point. A data audit plus a one-line config fix gained three.
Read more →A 600-million-parameter model that reads one message and returns five psychological state dimensions — and never writes a reply. It runs on a laptop in under a second, agrees with a frontier cloud scorer on 96.3% of clinical decisions (the v4 release announced here; the current v7 release scores 97.1% on its 453-turn final exam), and cost about seven dollars of API budget across the whole project. The weights are now public, because a component this small and this auditable should belong to the field, not to one company.
Read more →An interpretability paper found that language models keep a small, reportable 'workspace' above vast automatic processing — nobody designed it, it emerged from training. Held up against Hamo's own architecture, the match isn't a metaphor. It's the same structure, found three times, and it hands us a concrete list of things to fix.
Read more →A practicing therapist put it sharply: an Avatar bears no responsibility, because it has nothing to lose. The industry treats that as AI's nature. We treated it as a design problem — and gave every Avatar a finite, token-metered life: real metabolism, a ledger both sides can read, dormancy instead of death, and five red lines that keep the mechanism from ever becoming emotional blackmail.
Read more →An LLM asked to do therapy will improvise a technique based on vibes, and prompts don't hold — a sufficiently emotional message talks the model past its instructions. So we moved the decision out of the model. Pure Python adjudicates every turn; the LLM only scores state and writes words. Here's the architecture, the measured latencies, and what the rigidity cost us.
Read more →Most AI companions have no answer for the moment a user says they want to die. Hamo AI's protocol is documented, measured, and fail-closed: a three-stage detection cascade under 5ms for keywords, a fixed safety message that admits the AI's limits, a banner that can't be dismissed, a one-tap hotline, and an alert to the named therapist within seconds.
Read more →A private practice is capped by hours. An AI Avatar built and supervised by a therapist isn't — it extends their care into the 167 hours a week they're not in the room. Every client subscription splits into rough thirds: one to the therapist, one back to the client as tokens that keep their Avatar alive, one to the platform. Here's how it works, and what it isn't.
Read more →No — standard ChatGPT is not HIPAA compliant, because OpenAI does not sign a Business Associate Agreement for consumer plans. But 'not compliant' is the easy answer. The harder question is what a compliant AI system actually has to implement. Here are the seven technical controls, and an honest account of where Hamo AI stands.
Read more →Search for PHIPA-compliant AI in mental health and you'll find note-takers — tools that transcribe sessions for the therapist. Almost nothing addresses the harder category: AI that talks directly to clients between sessions. That's a different risk surface, and it needs different engineering. Here's the distinction.
Read more →Hamo doesn't practice one style of therapy — it practices nine evidence-based methods, from CBT to a newly released psychodynamically-informed exploration. But which method, how deep it goes, and whether it's safe to go there is never left to the model. Every method runs on one shared clinical spine. Here's the full range, and how Hamo chooses.
Read more →Most AI companions are optimized to keep you longer. Hamo's Therapist Mode is built to bring you closer to yourself. A deterministic, pure-Python clinical spine decides every turn — where you are right now, and what is allowed right now — before the model ever chooses its words. Here's how that differs from a chatbot, and an honest account of what's built versus what's roadmap.
Read more →Hamo AI is bringing the client experience into WhatsApp. Choose your Avatar therapist, connect your WhatsApp once, and start talking — no daily logins, no verification codes, no extra app to open. The same clinical depth and privacy protections, now in the chat you already use every day.
Read more →Hamo AI today unveils two foundational upgrades to its therapeutic AI engine: a 7-level Persistent Self model that builds long-term, cross-Avatar memory of who a person truly is, and AI Mind 2.0 — a 14-category narrative store that turns every conversation into structured understanding. Together they advance the Psychological Semantic Vector Space (PSVS) into a system that can truly Know U, Care U, and Enable U.
Read more →Hamo AI has completed a comprehensive PHIPA 6.1 technical compliance audit, implementing seven layers of privacy and security controls to protect patient health information — setting a new standard for trustworthy AI in mental health care.
Read more →On April 8, 2026, Hamo AI founder and CEO Chris Cheng pitched at the AI & Tech Networking event in Toronto, presenting the newly launched AI-Powered Therapist Avatar System to an audience of entrepreneurs, professionals, and investors.
Read more →A key milestone in protecting our brand in our home market.
Read more →On March 14, 2026, Hamo AI hosted its second product preview webinar for mental health professionals and officially released the first version of the HAMO AVATAR Psychological Semantic Vector Space Five-Way Benchmark Evaluation Report.
Read more →On February 27, 2026, Hamo AI hosted its inaugural free preview webinar, introducing the AI Avatar system to mental health professionals and demonstrating how therapists can create, own, and supervise their own AI-powered digital extension.
Read more →Hamo AI Technology Ltd., a Canada-based artificial intelligence company focused on human-centered psychology and relationship intelligence, today announced that it has officially joined the UBC Science Co-op Program as an approved employer partner.
Read more →Empowering Families with Emotional Intelligence Technology.
Read more →Strengthening Global Brand Presence.
Read more →On Oct 3, 2025, Hamo AI Secures Pre-Angel Investment to Accelerate Development of Family Relationship AI and Consumer App.
Read more →Hamo CMO Mona Xiao Meets with Empowered Startups VP Kyle Kirkegaard to Discuss Entrepreneurship in Canada.
Read more →