AI Assistance

Toggle

preset	Default	How to enable
basic	off	:features="{ ai: true }" + :ai-config
full	off	:features="{ ai: true }" + :ai-config
notion	on	Pass :ai-config (or demo env)

<YanivEditor preset="full" :features="{ ai: true }" :ai-config="aiConfig" />
<YanivEditor preset="notion" :ai-config="aiConfig" />
<YanivEditor preset="notion" :features="{ ai: false }" />

When AI is disabled, AI extensions and all AI UI entry points are unavailable.

Five Actions

After selecting text, trigger via the floating menu; with full preset and AI enabled, the header assistant row also shows AiMenuButton:

Action	Extension	Description
Continue writing	ContinueWritingExtension	Stream continuation
Polish	PolishExtension	Improve selected text
Summarize	SummarizeExtension	Extract key points
Translate	TranslationExtension	14 target languages
Custom	CustomAiExtension	User-provided prompt

UX: AiSuggestionPopover streams suggestions with accept/reject; AiHighlightMark highlights the suggestion region.

Config Sources (Priority)

getAiConfig() in client.ts is the single resolution entry point. It returns at the first level that matches, highest priority first:

1. Host-managed (instance-scoped) — the :ai-config prop, forwarded through the registry's getters (read fresh on every request)
2. getAiRequestConfig() — the registered host-managed copy, otherwise the localStorage user config (written by AiSettingsModal). It validates enabled and apiKey (except in proxy mode) and returns null when they don't check out
3. Host-managed fallback — when level 2 fails validation but a host config is registered, resolution still uses the host's provider / endpoint / model and does not fall through to .env (so an explicitly host-managed editor never silently picks up a build-time key)
4. Build-time — VITE_AI_* environment variables

Levels 2 and 3 look the host config up without an owner, so both fall through when several editors on the page pass :ai-config (see AI Config API — Multiple instances); requests made by the AI extensions rely on the instance-scoped getter at level 1 and are unaffected.

Upstream layers (registry getters, resolveAiExtensionOptions) only forward the host's raw values and never fill in defaults: without :ai-config, resolveAiExtensionOptions returns null and resolution falls through. Defaults (the provider's default endpoint / model, a 60s timeout) are filled in inside getAiConfig().

ai-config field changes do not trigger a session rebuild; changing model takes effect on the next request.

temperature / maxTokens are not part of ai-config and do not follow the levels above: their only source is VITE_AI_TEMPERATURE / VITE_AI_MAX_TOKENS (defaulting to 0.7 / 2048), and they apply whichever level supplied the key and endpoint.

With storageMode: "proxy" the key is held by your backend, so an omitted apiKey still counts as configured as long as there is a reachable endpoint.

Providers

openai | deepseek | aliyun | ollama | custom (OpenAI-compatible)

See AI Config API.

AI Subpackage

AI modules are imported from @yanivjs/yaniv-editor/ai; the root package does not re-export them:

import {
  ContinueWritingExtension,
  PolishExtension,
  SummarizeExtension,
  TranslationExtension,
  CustomAiExtension,
  AiMenuButton,
  AiSettingsModal,
  useAiConfig,
  createAiClient,
  AI_PROVIDERS,
} from "@yanivjs/yaniv-editor/ai";

Translation Languages

Simplified Chinese, Traditional Chinese, English, Japanese, Thai, French, Spanish, Portuguese, Korean, Vietnamese, Russian, German, Hindi, Indonesian.

Related

• AI Config API
• Integration Props