Ollama integrations

[browserstrangeness]

In the Agent Zero settings, one can use it locally via ollama with models such as llama3.2 and qwen-coder2.5 in different places. I have attempted to try non-openai text embedding models with no luck. Are there any existing models that have been tested and working, or for now is the openai text embedding the only one that is compatible so far? Thanks for reading.

[Javito3D]

Hi @browserstrangeness,
I’m going to try to help you in exchange for your help too. I’m not an expert in this, but I’ll explain how I managed to get it connected. I’m sure your level is much more advanced — personally, I have no idea what I’m doing, but I’m very interested in the topic and learning on my own.

Your issue is only with the embedding model, but I’m going to go over the full local setup to give you my personal example — as if I were writing this for myself:

---

🔧 How can you connect any embedding model to Agent-Zero locally without API keys and without OpenAI keys or models?

📝 I cloned the repository on my PC and used Docker to create the image and the container. Let’s go!

---

🧠 Required setup

To use models locally, you must select at least two different models:

• One for chat and utility
• One for embedding

Example:

• Use Ollama for chat and utility with model llama3.2
• Use LM_Studio for embeddings with model "Nomic Embed Text v1.5" (it's excellent)

---

🔌 Connecting models to Agent-Zero (local setup)

1. Edit .env file

In your bind-mounted host folder (/a0), open the .env file with your favorite IDE (e.g., VS Code) and add the model base URLs:

✅ Ollama

OLLAMA_BASE_URL="http://host.docker.internal:11434"

⚠️ Important:

• Do not use 127.0.0.1:11434 (that’s inside the container).
• Use host.docker.internal to auto-resolve the correct host IP.
• If you're not sure which port Ollama is using, run:

  ollama serve
  

  and check the terminal output.

---

✅ LM-Studio

LM_STUDIO_BASE_URL="http://host-url:port/v1"

⚠️ Tips:

• You can find the full URL in the “developer” section of LM_Studio when the server is running.
• You must include /v1 at the end for Agent-Zero to recognize the API correctly.

---

🔁 Restart the Docker container

After saving the .env file, restart the container to apply changes. These environment variables won’t apply on the fly.

---

2. Configure Agent-Zero Web UI

Open the Agent-Zero web UI, go to Settings, and set up the model providers and names:

Chat model:
  - Provider: Ollama
  - Name: llama3.2

Utility model:
  - Provider: Ollama
  - Name: llama3.2

Embedding model:
  - Provider: Lm_studio
  - Name: text-embedding-nomic-embed-text-v1.5@f32

---

⚠️ Linux users — model name warning

If your model name contains "@", replace it with "-" to avoid errors in Linux environments.

Wrong: text-embedding-nomic-embed-text-v1.5@f32  
Right: text-embedding-nomic-embed-text-v1.5-f32

---

3. No API keys required

No API keys are needed — both Ollama and LMStudio are open endpoints by default.

✅ If you prefer using the REST API of LMStudio (instead of the fake OpenAI API), use /api/v1 instead of /v1.

---

✅ Done!

You now have Agent-Zero running 100% locally with the models you want.

Now try to give me some information please because I`m gonna go crazy. Asking into an another task.

[Richard1185]

Embedding Model
Settings for the embedding model used by Agent Zero.
Embedding model provider
Select provider for embedding model used by the framework

Ollama
Embedding model name
Exact name of model from selected provider
deepseek-r1:8b

descarga el modelo en tu computadora y si te pide mas memoria pidele ayuda a copilot me funciono al 100% y no soy experto en la materia