An open-source SDK that lets weak and local LLMs work like frontier models on narrow workflows.
Macrokit is a runtime + SDK for shipping LLM applications under cloud-API constraints — data residency, compliance, air-gapped networks, or budget. It works by moving multi-step reasoning to design-time (a strong model encodes a workflow once as a deterministic macro) so the weak model only does intent classification at runtime ("user wants X → call macro Y").
The hard part for weak models is multi-step reasoning. The easy part is routing. Macrokit pushes the hard part offline so cheap/local models can do the routing reliably.
It recapitulates how brains manage the cost of thinking: a slow, expensive deliberation path (System 2 — the strong model) compiles repeated reasoning into a fast, cheap automatic path (System 1 — the macro), and the distillation gate compiles deliberation into reflex through repetition. A fast cheap reflex and a slow expensive mind, with the reflex carrying the load.
It's also an open format for macros — the deterministic, distilled form of a workflow. A skill tells a strong model how to think; a macro is the compiled result that runs on weak/local models. Macros can call MCP tools as primitives — Macrokit sits above MCP, not against it.
Launching publicly in Q3 2026 — see macrokit.dev for the launch essay and quickstart.
Macrokit deliberately keeps curated examples small. Vertical macro libraries live as standalone npm packages you publish yourself — see CONTRIBUTING_MACROS.md and the community registry for the listing process.
The registry is the seed of a broader ecosystem: a place to share and discover the vertical macro libraries authors build, so adopting a new vertical can start from proven macros instead of a cold start.
Built on a private production system that has been running this architecture in operations tooling for users without frontier-API access since early 2026.
Apache 2.0. See the LICENSE file in each repository.