Macrokit Studio

Macrokit Studio is an AI developer tools SDK that routes repetitive GitHub maintenance tasks to pre-authored macros using a tiny local model running entirely in the browser for developers and open-source maintainers. Rather than sending every request to a cloud API, Macrokit Studio separates two distinct phases: at design time, a strong model authors a macro — a recorded, reusable workflow — and at runtime, a small local model (as compact as Qwen2.5 0.5B) classifies incoming tasks and routes them to the matching macro, executing the work with no API key, no server, and no cloud cost. The demo ships with three pre-built macros targeting public GitHub repositories: triage_newest_pull() for inspecting and labeling the newest open pull request, summarize_open_issues() for listing open issues with labels and comment counts, and repo_health() for reporting stars, open-issue count, and default branch status. Model inference runs via WebGPU directly in the browser, and users can choose between Qwen2.5 0.5B (fastest), Qwen2.5 1.5B, or Llama 3.2 3B for better routing accuracy. A live local coverage meter shows what percentage of session requests are being served by macros rather than a cloud model. All macros operate in dry-run write mode against public repositories only — no accounts or private data are accessed. The open @macrokit/* SDK underpins the studio and is designed so that as novel tasks are handled by the strong model, the resulting macros accumulate in a library, progressively increasing the share of work served locally at zero marginal cost. Best for: open-source maintainers and developers who want to automate repetitive repo tasks locally without recurring API costs.

GitHub repo maintenance automation, pull request triage, open issue summarization, repository health reporting, local AI macro routing, developer workflow automation