SurfSense DocsOllama
Use models you run in Ollama
Ollama runs models on your computer and also serves them over the OpenAI API. SurfSense connects to that API like any other server.
You don't need Ollama to run models locally. SurfSense can download and run models itself. Connect Ollama if you already use it and want to keep your models there.
Start Ollama
- Install Ollama and pull a model, for example
ollama pull qwen3:8b. - Make sure Ollama is running. It listens on port
11434.
Connect it
Open Settings › Text gen and click Add model, then Use a server. In onboarding, click Connect under Use a server.
In Connect a server, enter a Name, such as Ollama.
Under Provider, pick Local or custom server. (Ollama Cloud is Ollama's hosted service, not the app on your computer.)
Enter the Base URL: http://localhost:11434/v1. Leave API key empty.
Click Save server. SurfSense asks Ollama for its models, then opens the list.
Click Use on a model. Test chat sends one short prompt so you can see it answer. Then click Use this model, or Use without testing.
Base URL
| Where things run | Base URL |
|---|---|
| SurfSense and Ollama on the same computer | http://localhost:11434/v1 |
| Ollama on another computer in your network | http://<host>:11434/v1 |
| SurfSense in Docker, Ollama on the computer running Docker | http://host.docker.internal:11434/v1 with Docker Desktop, or the computer's network address |
Replace <host> with that computer's IP address or name. Ollama only listens on its own computer by default. To reach it from another computer or from Docker, start it with OLLAMA_HOST=0.0.0.0.
Ollama has no sign-in, so then any computer that can reach the port can use your models. Limit who can reach it with a firewall, or put it behind a reverse proxy that asks for a password.
Include /v1. SurfSense uses the URL exactly as you enter it, and Ollama's OpenAI API lives under /v1.
Network permission
A server on the same computer (localhost or 127.0.0.1) needs no permission: nothing leaves your computer.
Any other address, including a computer on your own network or host.docker.internal, counts as sending data out. When you save, SurfSense asks Allow sending data to that address. Allow it to continue. You can turn it off later in Settings › Network.
Agentic mode and Ollama
Agentic mode needs a model that can use tools and a context window of at least 32,768 tokens. Ollama loads models with a short context window unless you raise it, for example with OLLAMA_CONTEXT_LENGTH=32768 when you start Ollama. With a short window, long Agentic tasks run out of room.
A model you run in Ollama shows as not tested for Agentic mode, even when SurfSense tested the full-size model through a provider. See Basic and Agentic chat.
Troubleshooting
The endpoint's model list could not be verified
SurfSense could not reach Ollama.
- Check Ollama is running.
- Check the URL ends in
/v1. - From the same computer, run
curl http://localhost:11434/v1/models. It should list your models.
No models listed
Pull a model with ollama pull, then open the server's group in SurfSense again. You can also type the model's name, such as qwen3:8b, under Exact model ID.
The first reply is slow
Ollama loads a model into memory on its first request. SurfSense waits up to five minutes for the first word.
Last updated on November 2, 2026