// the take
A Go CLI plus a library of markdown 'skills' that inject mobile-specific context (Android, iOS, Flutter, React Native, KMP) into AI coding assistants like Claude Code, Cursor, and Copilot CLI. For mobile teams already leaning on an AI assistant who want it to stop giving generic advice and actually know their platform's build quirks and conventions.
The host abstraction (cli/internal/host) is genuinely broad — it detects and writes config for a dozen+ assistants (Claude Code, Cursor, Codex, Gemini, Copilot CLI, Goose, OpenHands, etc.) instead of locking you into one vendor. The Go codebase is unusually well-tested for a CLI this size, with a _test.go next to nearly every source file. The skill packs are split per-platform (android/ios/kmp/flutter/react-native) so you install only what applies instead of one bloated bundle. The Graph component does real static parsing (dedicated Kotlin and Swift tokenizers/indexers, not just grep) to answer 'which files touch this symbol' before a fix.
The two technically interesting pieces — Graph (semantic indexing) and Brain (project memory) — are both still alpha, while the project's actual selling point today is a pile of markdown instructions whose quality you can't easily audit or regression-test. The roadmap's 'Agents' and 'Automated pipeline' phases (autonomous bug-fix-to-PR with no human in the loop) are vaporware right now, so don't evaluate this on what the README promises, only on the skills/CLI that exist. Value depends entirely on the AI assistant obeying the injected context well — if Claude Code or Cursor changes how they load project instructions, the whole mechanism can silently degrade. The CLI self-updates its own binary (mobiai update), which is convenient but means you're trusting a young, 489-star project's release pipeline on your dev machine.
We read the README, the directory tree and a key source file in October 2026. Nobody pays to be listed. How we judge →