Vvveb - AI Assistant Plugin

AI Assistant Plugin

Settings

Provider
All models.dev providers with plain key auth (list loads from the local catalog) plus custom for a local LLM (Ollama, LM Studio, llama.cpp) or any OpenAI-compatible server

API key
Key for the selected provider above (leave empty for a local server without auth)

Model
Provider specific, e.g. gpt-4o-mini, claude-sonnet-4-20250514, gemini-2.0-flash, deepseek-chat, llama-3.3-70b-versatile, llama3.1 (Ollama)

Advanced settings

Api mode
direct = browser calls the provider, proxy = requests are relayed through the admin backend to hide api keys from the browser. For a local server: direct reaches the LLM on your own computer, proxy reaches the LLM on the web server

Proxy url
Used when api mode is set to proxy

Session header
Upstream header carrying the chat session id on proxied requests (empty disables)

System prompt append
Sent with every request to set the assistant behaviour

Url
Optional endpoint override for the active provider above (e.g. a self-hosted gateway speaking its API, or a local server such as http://127.0.0.1:1234/v1/chat/completions for LM Studio); empty uses the catalog endpoint (custom defaults to Ollama at http://127.0.0.1:11434). Cleared automatically when you switch provider

Local presets (select the custom provider and point it at a loopback base; direct mode needs CORS allowed on the server, e.g. OLLAMA_ORIGINS for Ollama):

Temperature

Max tokens

Context size

Model context window in tokens (used for context usage & auto-compact, default 200k)

Image model

Image model for “Generate image from text” (OpenAI compatible, e.g. dall-e-3, gpt-image-1)

Image provider
Optional override for image generation (e.g. chat locally, images via OpenAI); empty uses the provider above. Local servers have no images API.

Image API key
Key for the image provider above; empty reuses the main API key

Monthly token cap
Max bulk tokens per calendar month and admin (0 = off); bulk actions fail fast when reached

Job webhook
POSTed as JSON when a bulk job finishes (own HTTPS endpoint recommended)

Image size

e.g. 1024x1024, 1792x1024, 1024x1792

Prompt playground

Try prompts against the active chat model before wiring them into jobs. Nothing is saved.

Template

Temperature

Max tokens

System prompt User prompt

Optional overrides for bulk translation (a cheap fast model saves money); empty inherits the provider, model and key above. Used by Bulk Translate and the write → translate pipeline.

Translate provider
Optional override for bulk translation (a cheap fast model saves money); empty uses the provider above

Translate model
Model for bulk translation; empty uses the model above

Translate API key
Key for the translate provider above; empty reuses the main API key

Translate glossary
One “Original = Replacement” per line; always used exactly in bulk translation

Translate reasoning
Off suppresses chain-of-thought on models whose catalog entry allows it (none/minimal); field translation needs none

Voice and covers for the bulk article writer. Empty inherits defaults; the image task settings above still drive cover generation.

Tone presets
One “Name | instruction” per line; selectable voice for the article writer

AI disclosure text
Appended to writer output when “Disclose AI” is ticked; empty uses a default sentence

Stock provider
Source for “Stock link” article covers: Auto tries each configured API key, then the keyless Unsplash search

Unsplash API key
Free tier at unsplash.com/developers (50 requests/hour); empty skips the API

Pexels API key
Free at pexels.com/api; empty skips the API

Write reasoning
Off suppresses chain-of-thought where the catalog allows it; outlines already carry the planning