OpenAI-compatible request shape
AIService builds the OpenAI chat-completions request. Point providerUrl at OpenAI, Anthropic, Together.xyz, Ollama, or Hugging Face Inference — no per-provider SDK in your code.
Features
Everything the AI layer needs, nothing the build layer does.
Anouk deliberately stops at the AI runtime and configuration. Build tooling stays with your framework of choice; the model plumbing is ours.
Provider layer
One adapter, every OpenAI-compatible endpoint.
Runtime provider switching
Provider URL, API key, model, and system prompt come from config that the user can change at runtime. Switch from OpenAI to a local Ollama endpoint without a rebuild.
Local-first friendly
Because the extension calls the provider directly, pointing at http://localhost:11434 (Ollama) keeps content on the user's machine end to end.
Runtime services
The plumbing that every AI extension re-invents.
Per-request response cache
Responses are cached by ${requestId}_${cacheKey}. Re-opening the same email or re-rendering the same sidebar returns the cached answer instead of repeating the spend.
Per-provider rate-limit queue
Calls are queued per provider so a burst of user actions does not stampede the API or earn a 429 mid-render.
AIService.call() entry point
AIService.call(instruction, content, requestId, cacheKey?) is the single call. It reads config, checks cache, queues, dispatches, decodes, caches, and returns a plain string.
UI + configuration
Give users a place to put their key.
Runtime settings panel
A bundled panel where the user supplies providerUrl, apiKey, model, and systemPrompt. Values persist in chrome.storage.local. Ship one bundle; users choose what to run against.
configManager
Config can also come from code via aiConfig.js — a default export with the same fields — useful for hardcoded development setups.
Framework-agnostic components
The UI pieces are plain DOM, so they drop into a Plasmo, WXT, or hand-rolled extension without a framework dependency.
Developer workflow
From zero to a loaded extension in one command.
anouk init scaffold
anouk init <name> writes a Manifest V3 project with aiService.js, configManager.js, settingsPanel.js, and a starter entry point — pre-wired and runnable.
anouk generate
Drop an extension, service, or config template into an existing project with anouk generate extension|service|config.
esbuild bundling
The template bundles src/* to dist/ with esbuild. Load dist/ unpacked at chrome://extensions and iterate.
Ready to wire it up?
The quickstart takes you from install to a loaded extension that talks to a model.