There's this thing I keep coming back to when I think about AI and creative work: the AI doesn't know what you actually want. It knows what you said. And those two things are often very different.
That gap — between what you meant and what the AI inferred — is what Canonic is built to close.
What Canonic Is
Canonic is a desktop world-building engine. You use it to define the rules, characters, locations, lore, and logic of a fictional universe — your Canon. Then, when AI generates stories, adventures, or characters inside that world, it's working from your definitions, not its own assumptions.
Think of it less like a writing tool and more like a constitution for a world. You're not writing the story. You're writing the rules the story has to follow.
The primary market I built this for is the homeschool community — families who want immersive, AI-powered educational experiences without handing the wheel entirely to the AI. A parent can build a historical Canon around ancient Rome, define the key figures, the political tensions, the daily life details, and then their kids can explore that world through interactive stories that stay true to what was actually set up. The AI generates within the boundaries. It can't freelance.
Secondary market: creative writers and world-builders who want the same discipline for their fiction.
The Stack
I went with Tauri for the shell — Rust-based, lightweight, cross-platform. The frontend is React + TypeScript. The backend is a Node.js sidecar (also TypeScript) that handles all the AI calls and database operations. SQLite via better-sqlite3 for local data storage.
The important thing about this stack is that everything runs locally. There's no cloud sync, no server, no account. Your worlds live on your machine. The only external calls are to the Anthropic API — and even that uses BYOK (bring your own key). You provide your own API key; I never touch it.
This was a deliberate choice. Canonic deals with creative work — often personal, sometimes deeply personal. That work shouldn't live on someone else's server.
The Module System
The core organizing concept is the module. Everything in a Canon is a module: characters, locations, lore entries, factions, plot points, events, items. Each module type has a defined field structure — personality, background, abilities for characters; terrain, atmosphere, history for locations; and so on.
Modules live inside zones (which are thematic groupings — Characters, Locations, Lore, etc.), and zones are part of a Canon. The World View shows you all your zones as orbs on a canvas, connected by labeled relationship lines. You can see at a glance how your world is structured.
When you export a Canon for Arc (the companion reading app), all of this — every module, every connection, every voice profile — gets packaged into a .canon file that Arc can import and use as ground truth for generation.
AI Models
I'm using three Claude models at different points in the workflow:
- Opus 4.6 for the Canon engine and Living World — the heaviest reasoning tasks
- Sonnet 4.6 for the Canon Builder, Adventure Mode, Keeper of Worlds, and Voice Profiles
- Haiku 4.5 for Character Chat — fast, cheap, and honestly great for conversational responses
All Anthropic, BYOK. I'm not interested in supporting multiple providers — at least not right now. Anthropic's models are the best fit for what Canonic does.
The Philosophy: Human-First, Always
If there's one thing that defines Canonic's design philosophy, it's this: the AI assists. The human decides.
Every AI-generated piece of content — whether it's a new module, a Living World event, or a Canon coherence suggestion — is a proposal. The user reads it, approves it, edits it, or discards it. Nothing enters the Canon automatically. Nothing becomes canon (lowercase) without explicit human confirmation.
This sounds obvious. It isn't. Most AI tools blur that line, and the result is that users stop feeling like authors and start feeling like editors of someone else's work. Canonic is built around the opposite experience: you're always the author. The AI is a very capable assistant who has read all the source material.
Where It's Going
Canonic is currently in beta polish. The core feature set is complete — Canon Builder, Character Chat, Adventure Mode, Living World, Keeper of Worlds, Voice Profiles, Canon Review, import/export. The next phase is refinement and then the Arc integration becomes the main story.
But the longer-term vision? There's a concept I'm sitting with called the Canon Metaverse — the idea that Canons can be linked as primary and secondary universes, with controlled "bleed-through" between them as stories deepen. The secondary world slowly emerges into the primary one. Things that break the first world's rules start appearing, because they're normal in the second. It's basically the Stranger Things Upside Down mechanic — and it turns out the test Canon I've been using is a perfect proof of concept.
That's not 1.0. But it's where this is going.
Building the entire world so that lotus flowers can bloom inside its walls. That's the best way I've found to describe it.