Seymour: a multi-tenant learning platform where AI stays an assistant
Every tenant gets its own brand, people, roles, and content, on one codebase. The software cracks nothing for you.
Organizations that want to own their learning platform instead of renting one, and want AI in it without handing the wheel to it.
A block-based editor with saved components, a color library, custom HTML in sandboxed iframes with scanning and audit logging, five baseline roles plus custom ones, and AI only where it helps: image generation with plain-language color naming, a writing assistant that edits for tone instead of rewriting, and a course assistant that walks ADDIE with you.
Still in development, and now mostly a reference project: what we learned building the editor, roles, and AI features shaped the school platform, which is being built fresh rather than on top of it.
Read the whole story
Seymour started as a question we kept hearing from organizations: why does owning a learning platform mean either renting one that isn't yours or building one from nothing? And underneath it a second question, more pointed in 2025: how do you put AI in a learning product without letting it do the learning for people?
The platform is multi-tenant on one codebase, so every organization gets its own brand, people, roles, and content without a separate deployment. Authors work in a block-based editor with saved components and a color library, so a course looks like the organization's course and not like the tool's. Custom HTML is allowed, but inside sandboxed iframes with scanning and an audit log, because the one thing a learning platform can't do is trust arbitrary code. Five baseline roles cover most organizations, and custom roles cover the rest.
The AI is where the philosophy shows. Image generation takes plain-language color names so a brand stays a brand. The writing assistant edits for tone and clarity instead of rewriting, so an author's voice survives. The course assistant walks an author through ADDIE, asking the questions an instructional designer would ask, and leaves the decisions where they belong. It helps you crack the egg; it doesn't crack it for you.
Seymour is still in development, and it has become mostly a reference project. Building its editor, its roles, and its AI features taught us what a school actually needs from a platform, and that knowledge went into the school platform above, which we chose to build fresh rather than on top of Seymour's generality. That's a good outcome for a lab project: the lessons shipped even where the code didn't.