LLM Index 0.8 + Ollama: "Could not find the GPT API server" while computing embeddings

Hello, I have issues with LLM Index and Ollam. Here is a Report:
Bug Report: XWiki LLM Index cannot compute embeddings with Ollama

Environment

  • XWiki: 18.4.0

  • Java: OpenJDK 21

  • Tomcat: 10.1

  • Database: PostgreSQL

  • Ollama: 0.30.10

  • Extensions:

    • LLM Application 0.8

    • LLM Models API 0.8

    • LLM Index API 0.8

    • Index for the LLM Application 0.8

Ollama Configuration

Server:

  • URL: http://127.0.0.1:11434/v1

  • No API token

  • Streaming enabled

Chat model:

  • Name: Mistral7b

  • Type: Large Language Model

  • Model ID: mistral:7b

  • Server: Ollama

Embedding model:

  • Name: all-minilm

  • Type: Embeddings Model

  • Model ID: nomic-embed-text:latest

  • Server: Ollama

Collection

Collection:

  • Company Knowledge

Configuration:

  • Embedding Model: AI.Models.all-minilm

  • Document Store: XWiki

  • Document Spaces: IT

The collection finds the XWiki documents correctly.

Problem

Re-indexing starts, but no chunks are created.

The collection shows:

Error computing the embedding:
RequestError: 500:
Could not find the GPT API server

What works

  • Chat with Ollama works correctly.

  • Mistral responds normally inside XWiki.

  • Direct REST call to Ollama works.

Embeddings API test:

POST http://127.0.0.1:11434/v1/embeddings

Body:

{
  "model":"nomic-embed-text:latest",
  "input":"Hello"
}

returns a valid embedding vector.

Therefore Ollama itself is working correctly.

Database checks

The embedding model is stored correctly:

serverName = Ollama
model = nomic-embed-text:latest
type = emb

The collection and imported documents also exist correctly in PostgreSQL.

Additional information

  • No related exception appears in the Tomcat logs when re-indexing.

  • The only visible error is:

Error computing the embedding:
RequestError: 500:
Could not find the GPT API server

Since chat works and embeddings work directly against Ollama, it appears that the indexer cannot resolve the configured GPT/Ollama server although the model references it correctly.

Is this a known issue with LLM Application / Index 0.8 and Ollama, or is an additional configuration step required?

Regards

Marco