Execution optimizer for AI-native code

migaki /mee-gah-kee/ [磨き] n. Japanese, from 磨く (migaku), "to polish"; polishing, refinement, honing.

Portable execution graphs, provider-aware lowering, and evidence traces. Migaki turns AI-native code into explicit plans, optimizes them for execution, and records what changed and why.

Migaki execution trace AI-native code enters Migaki as mIR, flows through context, cache, routing, verification, and evidence, then reaches an AI backend. Migaki passes over mIR AI-Native Code mIR plan Context dedup + trim Cache prefix layout Routing model + fallback Verification policy + validators Evidence trace bundle AI Backend
A portable plan becomes an optimized, evidence-backed run.

Project map

Work in progress, in one place.

Code TypeScript workspace, source, bootstrap, quality gate, and project direction. Wiki Architecture notes, decisions, and living documentation as the design sharpens. Backlog Implementation tasks, design follow-ups, defects, and milestone planning. Whitepaper Read the rendered paper, with a link back to the source repository.