Fosowl/agenticSeek · critical · Exception

Ollama connection refused at {host}. Is the server running?

Error message

Ollama connection refused at {host}. Is the server running?

What it means

This error is raised by ollama_fn (sources/llm_provider.py:198) when the HTTP request to the Ollama server fails with a 'connection refused'-style error that is not a plain httpx.ConnectError (that case is handled separately at line 188). The library detects the word 'refused' in the underlying exception string and re-raises it with the target host, meaning the TCP connection reached the machine but nothing is listening on the Ollama port. This is a server-side availability problem, not a client code bug.

Source

Thrown at sources/llm_provider.py:198

                model=self.model,
                messages=history,
                stream=True,
            )
            for chunk in stream:
                if verbose:
                    print(chunk["message"]["content"], end="", flush=True)
                thought += chunk["message"]["content"]
        except httpx.ConnectError as e:
            raise Exception(
                f"\nOllama connection failed at {host}. Check if the server is running."
            ) from e
        except Exception as e:
            if hasattr(e, 'status_code') and e.status_code == 404:
                animate_thinking(f"Downloading {self.model}...")
                client.pull(self.model)
                return self.ollama_fn(history, verbose)
            if "refused" in str(e).lower():
                raise Exception(
                    f"Ollama connection refused at {host}. Is the server running?"
                ) from e
            raise e

        return thought

    def huggingface_fn(self, history, verbose=False):
        """
        Use huggingface to generate text.
        """
        from huggingface_hub import InferenceClient
        client = InferenceClient(
            api_key=self.get_api_key("huggingface")
        )
        completion = client.chat.completions.create(
            model=self.model,
            messages=history,
            max_tokens=1024,

View on GitHub (pinned to ae57a23577)

Solutions

1. Start the Ollama server: run `ollama serve` (or start the Ollama desktop app) and verify it responds, e.g. `curl http://localhost:11434/api/tags`.
2. Check the host/port in config.ini: the address must point where Ollama listens (default port 11434); set OLLAMA_HOST if you use a non-default port.
3. If the app runs in Docker, use the docker-appropriate address (host.docker.internal or the host IP) instead of localhost, matching the library's internal_url handling.
4. If connecting to a remote Ollama server, confirm the server is reachable from your machine and the port is open (firewall/security group rules).
5. Restart the Ollama service if it crashed; check its logs for bind errors (port already in use, etc.).

Example fix

// before (config.ini)
server_address = localhost:11433

// after
server_address = localhost:11434  ; or start ollama with OLLAMA_HOST=0.0.0.0:11433

import socket

def assert_ollama_reachable(host="localhost", port=11434, timeout=2):
    try:
        with socket.create_connection((host, port), timeout=timeout):
            return True
    except OSError as e:
        raise RuntimeError(
            f"Ollama not reachable at {host}:{port} ({e}). Run `ollama serve` first."
        ) from e

assert_ollama_reachable()  # call before constructing the provider / ollama_fn

def is_connection_refused(err: BaseException) -> bool:
    """Narrow arbitrary provider exceptions down to the connection-refused case."""
    return isinstance(err, Exception) and "refused" in str(err).lower()

import time

def safe_ollama_fn(provider, history, retries=3, delay=1.0):
    for attempt in range(retries):
        try:
            return provider.ollama_fn(history)
        except Exception as e:
            if "refused" in str(e).lower() and attempt < retries - 1:
                time.sleep(delay * (attempt + 1))  # give the server time to start
                continue
            raise RuntimeError(
                "Ollama server is not running. Start it with `ollama serve`."
            ) from e

When it happens

Trigger: client.chat(model=..., messages=..., stream=True) is called against host (e.g. http://localhost:11434) and the underlying exception message contains 'refused' — i.e. no process is bound to that port. Note it is only reachable for exceptions that are not httpx.ConnectError but still carry 'refused' (some Ollama client/httpx versions wrap connect failures differently), and not for 404 responses (model not pulled), which are auto-retried via client.pull.

Common situations: Ollama daemon not started (forgot `ollama serve`); wrong host/port in config.ini (e.g. port 11434 vs a custom OLLAMA_HOST); running inside Docker where 'localhost' points at the container instead of the host (the library has an internal_url for docker, but a misconfigured server_address still refuses); server crashed or stopped mid-session; firewall/proxy rejecting the port on a remote server.

Understand the failure class

Background: ECONNREFUSED and "connection refused" / "could not connect to server" errors: what they mean and how to fix them — this error's family across 44 libraries.

• Connection failures: ECONNREFUSED, ECONNRESET, and friends — why connections get refused, reset, or dropped.

Related errors

• Ollama connection failed. is the server running ?
• Ollama connection failed at {host}. Check if the server is
• Cannot connect to LM Studio at {route_start} - check if serv
• Model not set
• Prompt file not found at path: {file_path}

AI-assisted analysis of Fosowl/agenticSeek@ae57a23577 (2026-08-30). Data as JSON: /api/errors/99ce4ccdc924add0. Report an issue: GitHub.