Concept · Thesis · 3 min read

One runtime for the whole desk

The typical AI-heavy desk is five tools taped together: a chat subscription, a terminal assistant, a local runner, a scheduler, and no shared memory, tooling, or audit trail between them. Kaptain's thesis is that this belongs in one local runtime.

The taped-together desk

Each tool has its own context, its own permissions story, its own bill, and no idea the others exist. The cost isn't just money — it's that nothing composes: the scheduler can't use the strong model's judgment, the terminal assistant can't see the project map, the local runner sits idle while the subscription burns.

One runtime

Kaptain is a single binary that serves the workbench, runs or connects every model route, carries the toolbelt, hosts the agents and their schedules, and records everything in one place — on your machine, inside a project boundary you chose. Every capability composes with every other because they share one runtime: a scheduled kapsule can delegate to a local model, over the mesh, with the same tools and the same audit trail as your live chat.

Model-agnostic, concretely

Not a slogan: four routes in one picker (CLI harnesses, Ollama, llama.cpp, cloud APIs), plus other machines' models over the mesh — swappable per chat, per kapsule, per fleet, without changing anything downstream of the pick.

Proof it composes

The Built with Kaptain shelf is the receipts: real applications planned, coded, and iterated end to end inside this runtime, on local hardware, by one person directing it.

Agent-readable version: index.md