AI writing assistance in the editor

Last modified: 2026-04-24 09:22av bruker: Erik Hagen ID: c5e5948f-2ac0-46f7-b27e-43a14f614046 ◍New SAMT-X/solution-samt-bu-docs

Users of the built-in editing interface can save time on writing, translation, and proofreading with AI directly in the editor — without SAMT-BU project incurring any API costs.

Concept

The user provides their own API key from an AI provider they already have an account with. The key is stored locally in the browser (localStorage) and used to call the provider’s API directly — no CF Worker or server-side component required.

Browser  →  [chosen AI provider's API]
             (key in localStorage, never sent to SAMT-BU infrastructure)

Supported providers (planned)

ProviderAPICORS from browserNotes
Anthropic Claudeapi.anthropic.com/v1/messages✅ with anthropic-dangerous-direct-browser-calls: trueclaude-haiku is cheapest
OpenAI / ChatGPTapi.openai.com/v1/chat/completions✅gpt-4o-mini = low cost
Google Geminigenerativelanguage.googleapis.com✅ via ?key=free tier available
Ollama (local)localhost:11434✅ (local)free, requires local install

UX in the TipTap editor

Three buttons in the editor toolbar plus a settings button:

[ Improve ]  [ Translate to EN ]  [ Shorten ]  [⚙ AI settings]

If no key is configured → clicking opens the settings dialog directly.

Technical implementation

All changes in custom-footer.html (theme):

  1. aiComplete(instruction, text) – shared function that delegates to the correct adapter
  2. One adapter per provider – wraps the call in the provider’s format, returns text (~20 lines per adapter)
  3. AI settings dialog – choose provider, enter key, choose model, test button; saved to localStorage
  4. AI toolbar in TipTap – three action buttons + settings button

localStorage keys

KeyContent
samtu-ai-provideranthropic / openai / gemini / ollama
samtu-ai-keyAPI key
samtu-ai-modelSelected model (e.g. claude-haiku-4-5-20251001)

Benefits

  • No cost to the project – user uses their own account
  • No new infrastructure – no CF Worker changes required
  • Extensible – new providers added as a single adapter with no structural changes
  • Works offline for Ollama users

Priority

Low — useful additional feature, but not critical to the core workflow.