AI & jiki

jiki is a natural fit for AI-powered development tools. A sandboxed browser runtime that can write files, install packages, and run code is exactly what an AI coding assistant needs to turn generated code into a live, running application.

This section covers two sides of the same coin:

Building AI apps with jiki — how to stream LLM responses, inject code into a container, and build live preview with error recovery
Using AI tools with jiki — how to give Claude, Cursor, Copilot, and other AI assistants the context they need to help you use jiki effectively

See it in action

jiki ships with three AI coding assistant examples, each demonstrating a different provider and use case:

Example	Provider	Model	What it builds
Claude Coding	Anthropic	Claude Opus 4.6	React components from natural language
OpenAI Coding	OpenAI	GPT-5.4 Mini	Astro sites with React islands
Mistral Coding	Mistral	Codestral	React components with live preview

All three follow the same architecture: stream tokens from an LLM API, extract code from the response, write it into a jiki container, and render a live preview in an iframe. The differences are in the API format and what gets generated.

Context files for AI tools

jiki provides structured context files so AI assistants can help developers use the library without hallucinating unsupported features:

File	Size	When to use
`llms.txt`	~0.5 KB	Spec-compliant index — points to the other files
`llm.txt`	~12 KB	Quick overview with skills and code snippets
`llm-full.txt`	~50 KB	Complete API reference, guides, and examples

Use llm.txt when you need a quick answer. Use llm-full.txt when you want the AI to help with implementation details across any jiki feature.

What’s in this section

Building an AI Coding Assistant — streaming, code injection, error recovery, and chat persistence patterns
Using jiki with AI Tools — how to configure Claude, Cursor, and Copilot for jiki projects
Provider Setup — API access for Anthropic, OpenAI, and Mistral in the browser