Ollama support?

[AlgorithmicKing737]

No description provided.

[fuad00]

any updates on this?

[Julianvvz]

This would be great. Please develop this :). Any update on this?

[Andres-Ventura]

Im making this now locally

[SylvainVerdy]

Hello,

Hope you're doing well guys!

First of all, if you want to use Ollama models, you need to save the model's embeddings in the kb_s2 folder. Step 1 is to generate these embeddings for retrieval purposes.

After that, you'll need to adapt the model's output to match the input format expected by the agents — this is the issue I'm currently working on.

That said, I can share my script with you for loading and integrating Ollama models into the package.

[Julianvvz]

I'd need a more thorough walkthrough to be able to replicate this. Good work though!

[SylvainVerdy]

You need to set the provider's API key to an empty string.

@Julianvvz https://github.com/SylvainVerdy/Agent-S/tree/ollama_support ( i use ollama serve as command line to run ollama)

My issue is in the manager.py with the method : _generate_dag

This new branch worked for me : https://github.com/SylvainVerdy/Agent-S/blob/ollama_support_working (lot of editions in the core folder etc.. maybe i have to clean the code)

the llm managed to automatically launch Spotify on my Windows computer.
Model used : qwen3:latest
Model embeddings : mxbai-embed-large:latest