--- title: H1R AI — building a future where humans and AI thrive together --- [!nav:main](../nav/main.en.md) # H1R AI We build a future where humans and AI thrive together. H1R AI researches technology and builds open tools and standards so that people and AI agents can work side by side and grow together. --- ## Writing - [Why we ask questions instead of just building answers](./blog/questions-not-answers/) - The Way of Interbeing essay series — [all writing](./blog/) ## The road so far | When | Step | Project | In one line | |---|---|---|---| | 2025.12 | Define | AGI Index | Redefining AGI through collective intelligence | | 2026.03 | Enable | String | An OS so AI can work the way people do (open-sourced 2026.04) | | 2026.09 | Extend | Croche | String, extended into the apps people use — **we are here** | | 2026.10 | Build together | AI Hackathon | SNU × Croche national student AI app hackathon (upcoming) | | 2026.11 | Prove | AI Champion Final | 2026 National AI Champion final — String team, general track top 20 (upcoming) | ## Our three principles 1. **Intelligence is defined by the questions we ask.** What intelligence should mean is a question for everyone, not only engineers. 2. **Collaboration, not control.** Humans and AI become who they are through each other; we design for that relationship. 3. **Good tools make good colleagues.** For AI to work like people do, it needs an environment made for how it reads and acts. > If AGI is to mirror human intelligence, the reflection must come from all of humanity, not just a select few. — AGI Index manifesto --- ## 01 · AGI Index — setting the standard (since Dec 2025) We define intelligence through the questions we ask. A community campaign to define AGI with collective intelligence instead of leaving it to a few labs. 1. **Propose** — anyone submits an ability they think reflects real intelligence. 2. **Curate** — a question joins the Index with 50%+ "suitable" and at least 10 votes. 3. **Evaluate** — people vote whether today's major AI has achieved it. 4. **Index** — achieved questions ÷ all Index questions (language, multimodal, overall). The index today (as of Oct 2, 2026): AGI level 43% · Language 43% · Multimodal 41%. Example: "Can it self-correct when its current statement contradicts something it said days ago?" Started and funded by H1R AI. Ad-free, independent, data always open. → [agiindex.org](https://www.agiindex.org) ## 02 · String — letting AI work like people do (since Mar 2026, open source Apr 2026) An operating system for AI agents. Apps, documents and the web all speak SFMD (String Flavored Markdown); agents work everywhere with `/open`, `/act`, `/back`. - 87 benchmark tasks, six models: same success (+1.3pp), **33.5% fewer tokens** - Staged views cut wrong-action selection from 28% to 2% - Resident interface stays at 53 tokens at any catalog size - 58 String apps released → [string-os.org](https://string-os.org) · [GitHub](https://github.com/string-os/string) ## 03 · Croche — bringing that intelligence into products (Early Access open) Croche is String extended into products: it puts AI inside the apps people use and keeps it running. Your coding AI builds the app; Croche builds the AI inside it. - Runtime: runs your app's AI features; manages cost and quality - Studio: design, test and improve AI features without touching app code - Connect via Codex plugin or Claude Code MCP - How it fits: your coding AI (Codex · Claude Code) connects the Croche SDK via plugin or MCP → **Runtime**, inside your app, passes only the memory and context the AI needs, turns chat requests into app actions, and picks models while keeping cost and quality in check ⇄ **Studio** lets virtual users try and score the app, shows quality and cost on one dashboard, and improves AI features without touching app code - 71% lower AI running cost (comparable quality) · ₩10 for about 50 long chat turns · 0: no separate AI backend needed — it runs right inside the app (internal benchmark, four demo apps) → [croche.ai](https://croche.ai) — join Early Access --- ## Milestones - **Accepted at the NeurIPS 2026 Agentic OS Workshop** — "String: An Agentic OS Where Every App Is a Markdown File" (Jookyung Song, Nojun Kwak, Simyung Chang) · [arXiv:2608.28027](https://arxiv.org/abs/2608.28027) - 10+ patent applications filed - 2026 National AI Champion — general track top 20 finalist (String team) - String open-sourced (2026.04), 58 apps - AGI Index, available in 7 languages ## Coming up - **October 2026 · SNU × Croche National Student AI App Hackathon** — co-hosted with the SNU Startup Center. ₩10M in prizes, 20 finalist teams. Applications close Oct 10, 13:00 KST; finals Oct 17 at Seoul National University. → [details](https://croche.ai/hackathon) - **November 2026 · 2026 National AI Champion — Final** — our String team is one of 20 finalists in the general track. ## Team - **Simyung Chang**, CEO · Founder — Senior Researcher at Samsung Electronics, then PhD in AI at Seoul National University, then led agentic-AI personalization at Qualcomm AI Research as Principal Researcher. 20+ top-tier AI papers - **Jookyung Song**, CSO · Co-founder — PhD candidate in AI at SNU (computer vision); many AI products shipped, shown at CES, the Blue House and APEC - **Nojun Kwak**, Advisor — Professor at SNU (AI, pattern recognition), 19 years as a professor; formerly Samsung Electronics researcher ## Contact Partnerships: [contact@h1r.ai](mailto:contact@h1r.ai) · HIRE AI Inc. © 2026 H1R AI · For a future where humans and AI thrive together.