Hidden Door is a Brooklyn-based interactive-story platform founded by Hilary Mason (a well-known computer scientist and four-time founder) and Matt Brandwein, built on the thesis that "any story can become a game" with generative AI. Its pitch — "Step into the stories you love, or dream up a world of your own" — is delivered through two halves. The consumer side is play: you pick a door into an officially licensed world (the homepage shows Bram Stoker's Dracula, Conan Doyle's Sherlock Holmes, L. Frank Baum's The Land of Oz, Lovecraft's The Call of Cthulhu, Jane Austen's Pride & Prejudice, Charles Stross's Saturn's Children, and Pressman's The Crow), then play through it as interactive fiction, chat in-character with persistent NPCs ("he remembers what you told him last night"), host social games with friends, or follow the world as microdramas, comics and audio series. The creator side is Atlas, a world-building studio where authors and rights-holders model a universe's characters, physics, lore, rules and canon, then tune and test it; the platform claims a rights-holder can let fans play in their world while keeping control of what is canon and what may change. Under the hood Hidden Door emphasizes it is not a wrapper on a general LLM: it describes a typed, auditable narrative world model with embeddings, an agentic story engine that turns each player intent into beats and world-state changes checked against canon before display (e.g., "Dracula must be invited in — he stops at the threshold"), roughly 100 specialized narrative tasks, and fine-tuned small models that — on one core task — reportedly beat a 120B general model. It also ships a Creator MCP so worlds can be built/extended from Claude and other assistants, plus an Experience API. The company has raised ~$7M from Northzone, Betaworks, Makers Fund, Brooklyn Bridge Ventures and Homebrew, and its CEO has publicly presented the company's IP-to-game vision at industry conferences.

Core Highlights

  • Step into officially-licensed worlds (Dracula, Sherlock Holmes, The Land of Oz, The Crow, Cthulhu…) or build your own in the Atlas world-builder.
  • Multiple play formats from one world model: interactive fiction, persistent character chat, multiplayer social games, microdramas, comics, audio series.
  • Agentic "story engine" enforces canon/rules on every turn (intent → plan → canon check → world-state update).
  • Not a generic-LLM wrapper: purpose-built narrative world model + fine-tuned small models (claims 1.7B beats 120B on one task).
  • Creator MCP + Experience API; rights-holders set canon boundaries so IP owners keep control.
  • Pros
  1. Elite, verifiable founder + institutional backing. CEO Hilary Mason is a genuinely prominent AI figure (four-time founder, former Distill Research lead) with a named public keynote (T-EDGE 2023) explaining the product; Tracxn/CB Insights record ~$7M from Northzone, Betaworks, Makers Fund, Brooklyn Bridge Ventures and Homebrew. This is far above the self-published one-pagers in your recent set.
  2. Real, licensed IP is a hard-to-copy moat. Officially tagged worlds (Stoker's Dracula, Conan Doyle, Oz, Cthulhu, Austen, Stross, The Crow) are exactly what most AI-fiction apps cannot legally offer — the platform's whole "respectful of IP; rights-holders set canon" design targets this.
  3. Correctly solves the two hardest problems in AI fiction — coherence and canon. A typed world model plus a canon-checking agentic engine addresses long-session drift and lore-contradiction that plague naive "LLM dungeon master" products.
  4. Small-model efficiency story is credible and cost-relevant. Fine-tuned 1.7B-outperforming-120B claims (self-reported but specific) map to serving millions of interactive sessions economically — a real inference-cost differentiator.
  5. One world, many formats. Turning a single world model into interactive fiction, character chat, social play, microdramas, comics and audio multiplies monetization surfaces from one asset — attractive to rights-holders.
  6. Creator MCP + Experience API extend reach. Letting Claude-class assistants build/extend worlds and exposing an experience API is an early, forward-looking surface most narrative apps lack.
  7. Cons
  8. Consumer-review footprint is essentially absent. No G2/Capterra/Trustpilot corpus; adoption, retention and content quality cannot be judged from independent reviews — only from the founder's pedigree, funding and the IP wall.
  9. Small team / key-person + scale risk. ~8 employees and a $7M seed-stage balance sheet against a heavy vision (licensed IP across film/TV/books + a custom model stack + multi-format delivery) is an ambitious surface area for a small company to sustain.
  10. Performance claims are self-reported. "1.7B beats a 120B model on a core task," "~100 specialized narrative tasks," and IP-license breadth come from the company's own site; no third-party benchmark or named license term is published.
  11. Pricing is not published in static copy. The site has a Pricing page but exact plan/credit figures are not in the retrievable HTML — cost must be confirmed on the live app before quoting (do not invent tiers).
  12. Licensed-world catalog may be narrow / subject to licensing churn. The "official" IP shown depends on active rights agreements; the depth and longevity of the world library (versus user-generated worlds) are not documented.
  13. Heavy IP/licensing posture cuts against open creativity. "Rights-holders set what is canon and cannot change" is great for brands but constrains the freeform fan-fiction freedom that made some rival AI-fiction communities popular — some users may find worlds feel gated/railroaded by canon.
  14. Who Is It For
    Best suited for fans of licensed fiction and tabletop/interactive-fiction players who want to roleplay inside worlds they already love (Dracula, Sherlock Holmes, Oz…) with a coherent, canon-aware AI narrator, and for authors, filmmakers and IP rights-holders who want to turn an existing universe into a playable, controllable experience (via Atlas + Creator MCP) without commissioning a multi-year game adaptation. Not suited for buyers needing transparent published pricing before trialing (pricing not in static copy); users wanting a verifiable, heavily-reviewed consumer platform with published metrics (none exists); people wanting fully open, ungoverned freeform fan-fiction (canon is deliberately constrained); teams expecting a general-purpose writing assistant (it is a play/world platform, not a doc tool); and anyone assuming a large, established vendor — it is a small (~8-person) seed-stage studio.