Odysseus AI
Back to troubleshooting

Odysseus AI Ollama Connection Error

Last updated: June 6, 2026

If Odysseus shows connection refused, no models, or a failed Ollama endpoint test, check where Ollama is running and which network address Odysseus can actually reach.

Quick answer

If Odysseus runs in Docker and Ollama runs on the host machine, do not use http://localhost:11434 inside Odysseus. Use http://host.docker.internal:11434/v1 on Docker Desktop, or the host IP plus /v1 on Linux.

1. Confirm Ollama Works by Itself

Run these on the machine where Ollama is installed:

Copy command
ollama list
ollama pull llama3.1:8b
curl -fsS http://localhost:11434/api/tags

If this fails, fix Ollama before changing Odysseus. If it works, continue with the Docker or network endpoint checks below.

2. Start Ollama on a Reachable Address

Ollama often listens only on loopback. For a Docker container to reach host Ollama, start Ollama on an address reachable from Docker.

Copy command
OLLAMA_HOST=0.0.0.0:11434 ollama serve

Keep this on a trusted machine or private network. Do not expose unauthenticated Ollama to the public internet.

3. Use the Right Endpoint for Your Setup

Odysseus native, Ollama same machine

http://localhost:11434/v1

Odysseus in Docker Desktop, Ollama on host

http://host.docker.internal:11434/v1

Odysseus in Linux Docker, Ollama on host

http://<host-ip>:11434/v1

Ollama on a private remote machine

http://<private-ip-or-vpn-name>:11434/v1

4. Test from Inside the Odysseus Container

If the host check works but Odysseus still fails, test the endpoint from the same Docker network as the app.

Copy command
docker compose exec odysseus curl -fsS http://host.docker.internal:11434/api/tags

On Linux Docker, replace host.docker.internal with the host IP address reachable from the container.

5. Fix No Models Listed

A reachable Ollama server can still return an empty model list. Pull a model that fits your machine, then refresh the provider in Odysseus.

Copy command
ollama pull llama3.1:8b
ollama list

For low-memory machines, start with smaller quantized models and review the hardware requirements guide.

6. Map the Error to the Fix

Connection refused

Ollama is not listening where Odysseus is trying to connect, or Docker is using the wrong host address.

404 or provider test failed

The endpoint may be missing /v1 when Odysseus expects an OpenAI-compatible API.

No models available

The Ollama server is reachable but has no pulled models, or Odysseus is pointed at a different Ollama instance.

Timeout

The endpoint is blocked by firewall, VPN, Docker network rules, or a remote host that is not reachable from the app container.

Verify Against Official Docs

Endpoint field names and provider UI details may change as Odysseus evolves. For the latest settings screen and supported backend names, check the official GitHub repository.

FAQ

Why does Odysseus fail to connect to Ollama?

The common cause is using localhost from inside Docker. A container's localhost is the container itself, not the host machine running Ollama. Use host.docker.internal on Docker Desktop or the host IP on Linux.

Does the Ollama endpoint need /v1?

Use the /v1 suffix when Odysseus expects an OpenAI-compatible API endpoint. For Docker Desktop, that usually means http://host.docker.internal:11434/v1.

Why are no Ollama models listed in Odysseus?

Ollama may be reachable but empty. Run ollama list and pull at least one model, such as llama3.1:8b, before selecting a model inside Odysseus.

Can Ollama run on another machine?

Yes. Use the other machine's reachable private IP or VPN address, keep the /v1 suffix when required, and avoid exposing Ollama directly to the public internet.

Should I expose Ollama publicly to fix connection refused?

No. Keep Ollama on localhost, a private LAN, VPN, or reverse proxy with access control. Public unauthenticated Ollama endpoints are unsafe.

Installation support

Need hands-on installation support?

Setup Helper is the self-service planner. Installation Support is for users who want a human to review the route, diagnose logs, or get Odysseus running. We confirm scope and price before any payment.

Request Log Diagnosis

$49. Pay only after we confirm the scope.

Selected package

Best for

  • Docker, Ollama, port 7000, admin login, or GPU checks already failed.

You get

  • One focused log review
  • Likely root cause
  • Exact next checks
  • Credit toward Remote Install

Not included

  • Full remote install
  • Hardware driver surgery
  • Guaranteed large-model performance
Please do NOT send passwords or API keys.

Related Guides

Ollama + Models

Connect Ollama, OpenRouter, vLLM, or another model backend.

Docker Setup

Run Odysseus in Docker and connect it to host services safely.

GPU Not Detected

Fix Docker GPU passthrough before tuning local model performance.

Hardware Requirements

Choose a model that fits your RAM, VRAM, and storage.

Windows Guide

Windows-specific Docker Desktop, WSL2, and Ollama endpoint notes.

Troubleshooting

Return to the main setup issue checklist.