AI Providers

The server uses OpenAI-compatible HTTP APIs for both embeddings and chat completions. Provider selection is based on config.

Embedding providers

Selection logic (current code):

• provider: "ollama" -> Ollama embeddings
• provider: "chutes" -> Chutes-compatible embeddings
• otherwise:
  • if models.embedder.models[0] is set -> OpenAI-compatible embeddings
  • else -> BGE-compatible embeddings

Config fields:

• models.embedder.endpoint
• models.embedder.token
• models.embedder.models (first entry is used)
• embedding.dimension, embedding.max_batch, embedding.max_chars, embedding.rate_limit_tpm

Chat completion providers (Ask AI)

/api/askai reads model settings from the config file and environment variables:

• CHUTES_API_URL (endpoint)
• CHUTES_API_MODEL
• CHUTES_API_TOKEN

If these are unset, it falls back to values in the config file (models.generator).

Reranker (optional)

If models.reranker.endpoint is configured, the retrieval pipeline will rerank candidates using the configured endpoint.