feat: support custom OpenAI-compatible API endpoints

[Eistee82]

Summary

• Add configurable base URL in adapter settings to support any OpenAI-compatible API provider (Ollama, LM Studio, Google Gemini, DeepSeek, OpenRouter, etc.)
• Models are fetched dynamically from the configured API endpoint instead of a hardcoded list
• Add "Test API connection" button in adapter settings (server-side via sendTo)
• All API calls (model listing + chat completions) are proxied server-side to avoid CORS issues with local providers
• Strip LLM thinking artifacts (<think>, <|endoftext|>, etc.) from responses for local models
• Error handling with user-friendly messages and retry button for failed model loading
• Translations for all new UI elements in 11 languages
• Documentation in English and German with recommended providers (Google Gemini free tier, DeepSeek, local models)

Changes

• src/main.ts — Add testApiConnection and chatCompletion sendTo handlers
• src-editor/src/OpenAi/OpenAiDialog.tsx — Refactor to use server-side sendTo instead of browser-side OpenAI SDK, add dynamic model loading, error handling, response cleanup
• admin/jsonConfig.json — Add base URL field and test connection button
• io-package.json — Add gptBaseUrl native config
• admin/i18n/*.json — Translations for 3 new admin UI keys (11 languages)
• src-editor/src/i18n/*.json — Translations for 8 new editor UI keys (11 languages)
• docs/en/README.md, docs/de/README.md — Full documentation with provider recommendations

Test plan

• Configure OpenAI API key only (no base URL) — should work as before
• Configure Google Gemini (base URL: https://generativelanguage.googleapis.com/v1beta/openai) — models load dynamically, code generation works
• Configure Ollama (base URL: http://localhost:11434/v1) — models load, code generation works
• Test "API connection test" button with valid and invalid configs
• Test with unreachable endpoint — error message + retry button shown
• Verify all 11 language translations render correctly in admin UI

🤖 Generated with Claude Code

[Skeletor-ai]

Bug: Empty API key rejected for local providers (Ollama, LM Studio)

When using a local provider like Ollama that doesn't require authentication, the adapter currently rejects the request with 'No API key provided' before even attempting the connection. This means the model list stays empty and no scripts can be generated.

Affected code in build/main.js (both testApiConnection and chatCompletion handlers):

if (!apiKey) {
    this.sendTo(obj.from, obj.command, { error: 'No API key provided' }, obj.callback);
    break;
}

Workaround: Users can enter any dummy string (e.g. ollama) as the API key — Ollama accepts any Bearer token value.

Suggested fix: Allow empty API key when a custom baseUrl is set (i.e. not the default OpenAI endpoint):

if (!apiKey && !baseUrl) {
    this.sendTo(obj.from, obj.command, { error: 'No API key provided' }, obj.callback);
    break;
}

Tested with Ollama http://192.168.0.141:11434/v1 — works fine with dummy key, fails silently (empty model list) when key is empty.

[GermanBluefox]

@Eistee82 Have you tried Models? E.g. babbage-002 delivers: This is not a chat model and thus not supported in the v1/chat/completions endpoint. Did you mean to use v1/completions?
There are some models like sora or Dalle, the are definitely not for code generation

[GermanBluefox]

@Eistee82 Thank you for updating AI agents. The version is now in build.
You talked about AI-Chat. Can we discuss that somehow?

[Eistee82]

@Eistee82 Thank you for updating AI agents. The version is now in build. You talked about AI-Chat. Can we discuss that somehow?

https://t.me/AMEistee