AI Coding Support

When an AI coding agent works in a raw codebase, it falls back on its training data - which is months out of date and knows nothing about your project's conventions. The result is code that compiles but picks the wrong seam: overriding a unified method instead of writing a custom one, hardcoding a value the inheritance system should own, logging with console.log instead of the Alokai logger.

Alokai ships @alokai/ai-toolkit to close that gap. It delivers AI context - agent skills plus AI-tagged documentation - straight into your project through npm, so the agents you already use answer with the conventions the framework actually enforces.

What you get

@alokai/ai-toolkit is a versioned npm package that ships with every Alokai storefront as a dev dependency. It carries three things:

• Skills - focused, task-scoped guides that an agent loads on demand when it recognizes a matching task, such as middleware caching, multistore, theming, or i18n.
• Routing instructions for AGENTS.md - a managed block that tells any agent how to find and use the rest of the context.
• Version-matched docs - pages from the Alokai documentation site bundled into the package, so agents read documentation pinned to the version you have installed instead of whatever their training data remembers.

Because it's a package and not a one-time file copy, the context evolves with the product: upgrade the toolkit and your agents pick up the latest conventions on the next install.

What it looks like in practice

Ask your agent to add a Spanish locale to the storefront. Without the toolkit, the agent improvises: it hardcodes strings, invents its own translation setup, and misses the per-locale currency wiring. With the toolkit, the agent recognizes the task, loads the i18n skill, and follows the same steps an Alokai developer would - creating the lang/es/ translation files, registering the locale, and configuring the currency the way the framework expects.

The same applies across the framework: theming, multistore overrides, custom middleware methods, logging. The agent stops guessing and starts following the conventions your project is built on.

More than code generation

The same context that guides code changes makes your agent a reliable way to learn the framework. Ask it questions the way you'd ask a senior Alokai developer:

• "How do I expose a custom field from the platform response on the unified product?"
• "How does store inheritance work in this project?"
• "What are all the places I need to touch to add a new locale?"
• "Why does this project forbid console.log and what should I use instead?"

The answers come from the version-matched docs and skills installed in your project, so they describe the framework as it works in your repository - not as your agent's training data remembers it.

Already in your project - or one command away

Newly generated Alokai storefronts include @alokai/ai-toolkit out of the box, fully synced and ready. If your project was generated before the toolkit existed, adopting it takes a single command:

yarn alokai ai sync

Confirm the two prompts: the CLI installs the package at the version matching your ecosystem release, then adds itself to scripts.postinstall so the context stays in sync on every install.

Either way, you can verify what your agents see by listing the synced skills:

ls .claude/skills   # Claude Code (ai-toolkit-* entries)
ls .agents/skills   # Codex, Gemini CLI, GitHub Copilot

Explore