This page is machine-translated from French. Read the French original.

Supported providers

The exact list of providers available in your version is the one displayed in the settings screens. This page provides an overview for version 1.17.2.


Language models

Local execution, managed by WivenLLM

Integrated engine: a catalog of open models that the application downloads and runs itself, without an external service. This is the only case where WivenLLM manages the model lifecycle itself.

Local inference servers

The most popular self-hosting solutions are supported — both desktop and server template servers, including runtimes offered by hardware manufacturers: notably Ollama, LM Studio, LocalAI, NVIDIA NIM, and Microsoft Foundry Local. WivenLLM connects to them as a client.

Special case: any server exposing an OpenAI-compatible API can be declared, even if it is not specifically listed, using the generic connector. Simply provide its address.

Commercial Suppliers

Major model publishers are supported, as well as cloud platforms from leading hosting providers and specialists in rapid inference: including OpenAI, Azure OpenAI, Anthropic, Google Gemini and Vertex AI, AWS Bedrock, Cohere, DeepSeek and Groq.

Aggregators

Gateways providing access to many models behind a single key are also supported, allowing you to change models without changing configuration: notably OpenRouter and LiteLLM.

European and Swiss accommodation

European providers, including a Swiss hosting company (Infomaniak AI), are among the integrated solutions, such as Mistral—useful when a fully local solution isn't feasible but the location of processing is important. Verify the processing location with each provider.


Embedding engines

Integrated (by default) — runs locally, without a key. The template is downloaded once on first use, then nothing is downloaded after that. Default template designed for English; a multilingual template is available.

Local — Local inference servers exposing an embeddings service can be used: notably Ollama, LM Studio and LocalAI.

External — the main embedding providers on the market are supported (including OpenAI, Azure OpenAI, Gemini, Cohere, Mistral AI and Infomaniak AI), as well as any service exposing an OpenAI compatible API.

Reminder : Changing the embedding engine resets all indexes and requires re-adding documents to each space. → Vector base and embeddings


Vector bases

Integrated local database (default) — stored on disk, no infrastructure to operate.

Self-hostable databases — including Qdrant, Chroma, Weaviate, Milvus and PGVector, the vector extension of PostgreSQL, which allows the reuse of an existing database.

Managed services — including Pinecone and Zilliz Cloud, as well as the cloud offerings of the above solutions.

A managed service means that your vectors and their metadata are outside your scope.


Image generation

Local — an integrated engine managed by WivenLLM, as well as the most widely used self-hosted image generation servers: notably Automatic1111 and LocalAI.

External — the main commercial image generation services: including OpenAI, Gemini, Stability AI and Infomaniak AI Tools.


Video generation

Local — integrated engine, or self-hosted ComfyUI chain (LTX-2).

This function is significantly more demanding in terms of hardware and time than image generation.


Text-to-speech

  • Browser voice (default) — no configuration, nothing comes out of the WivenLLM server.
  • High-quality local voice, played in the browser.
  • Self-hosted synthesis engine.
  • External speech synthesis services, including OpenAI and ElevenLabs, as well as any service exposing an OpenAI-compatible API.

Audio transcript

For imported audio and video files:

  • Integrated engine (default) — runs locally, keyless, template downloaded on first use.
  • External transcription services: OpenAI and any OpenAI compatible service.

Voice recognition

For the dictation exercise in the chat:

  • Whisper locally (by default).
  • Browser recognition, which usually transmits the audio to the browser editor.
  • A recognition engine on your local server.
  • External services: OpenAI, Deepgram, Groq and any OpenAI compatible service.

Search engines for agents

The web search capability, which can be enabled, relies on a search engine of your choice. Major search API services are supported (including Brave Search, Tavily, Exa, SerpApi, Bing Search, and Perplexity), as well as DuckDuckGo and the self-hosted SearXNG engine—allowing you to maintain web search functionality without relying on a commercial service. If no engine is selected, queries are routed through a default public search engine.


Choosing: the question to ask yourself

For each brick, a single question determines the choice:

Do I accept that this data is outside my scope?

Brick What comes out with an external supplier
Language model The messages and the documentary context
Embeddings engine The text of all indexed documents
Vector base Vectors and their metadata
Transcription Audio files
Text-to-speech The text of the answers
Image generation The descriptions
Web search The requests

The embeddings engine line is the one that is most often forgotten, and it is the one that exposes the most content.