Skip to main content

opencode use remote ollama

Running OpenCode with a Remote Ollama Server

I recently changed my local AI development setup so that OpenCode no longer runs models on the same machine where I edit code. Instead, OpenCode connects to a remote Ollama server over my local network.

This approach lets me keep my development environment lightweight while dedicating another machine to model inference.

Why use a remote Ollama server?

Running Ollama remotely offers several advantages:

  • The development machine remains responsive while the model generates responses.
  • GPU resources can be centralized on a dedicated machine.
  • Multiple computers can share the same inference server.
  • Updating or changing models only needs to be done on one system.
  • The OpenCode configuration remains simple.

As long as the network latency is reasonable, the experience is very close to using a local Ollama instance.

My OpenCode configuration

OpenCode supports providers compatible with the OpenAI API. Since Ollama exposes such an interface, configuring a remote instance is straightforward.

My configuration looks like this:

{
  "$schema": "https://opencode.ai/config.json",
  "model": "jaahas/qwen3.5-uncensored:4b",
  "provider": {
    "ollama": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Remote Ollama",
      "options": {
        "baseURL": "http://192.168.122.1:11434/v1"
      },
      "models": {
        "jaahas/qwen3.5-uncensored:4b": {
          "name": "Qwen 3.5 Uncensored 4B",
          "tools": true
        }
      }
    }
  }
}

The important part is the baseURL option, which points OpenCode to the remote Ollama server instead of localhost.

The model

For this setup I use the following model: 

jaahas/qwen3.5-uncensored:4b

This 4-billion-parameter model is a good fit for my hardware. It provides useful coding assistance while remaining small enough to run comfortably on a GPU with limited VRAM.

I also enabled tool support: 

"tools": true

This allows OpenCode to invoke the tools it needs during interactive coding sessions.

Things to keep in mind

  • Configure Ollama to listen on the network interface instead of only on localhost.
  • Open port 11434 on the server if a firewall is enabled.
  • If the server is exposed outside a trusted network, place it behind a reverse proxy and add authentication rather than exposing Ollama directly.
  • Verify that the remote machine has enough GPU memory for the models you intend to run.

Final thoughts

Separating the development environment from the inference server has worked well for me. OpenCode behaves as if the model were local, while the computational work is handled by another machine.

If you already have a workstation or home server capable of running Ollama, configuring OpenCode to use it only requires changing the provider configuration. It is a simple way to make AI-assisted development available from multiple computers without duplicating model installations.

Popular posts from this blog

Undefined global vim

Defining vim as global outside of Neovim When developing plugins for Neovim, particularly in Lua, developers often encounter the "Undefined global vim" warning. This warning can be a nuisance and disrupt the development workflow. However, there is a straightforward solution to this problem by configuring the Lua Language Server Protocol (LSP) to recognize 'vim' as a global variable. Getting "Undefined global vim" warning when developing Neovim plugin While developing Neovim plugins using Lua, the Lua language server might not recognize the 'vim' namespace by default. This leads to warnings about 'vim' being an undefined global variable. These warnings are not just annoying but can also clutter the development environment with unnecessary alerts, potentially hiding other important warnings or errors. Defining vim as global in Lua LSP configuration to get rid of the warning To resolve the "Undefined global vi...
Read more

LazyGit AI Commit Message

Having AI‑generated commit messages directly integrated into LazyGit If you use LazyGit every day, you already know how it turns Git from a chore into something you can actually enjoy. But there is one part of the workflow that still tends to feel a bit tedious: writing good commit messages. In this post, I show how to plug OpenAI models directly into LazyGit using a tiny one‑file BASH script, so you can get AI‑generated commit messages based on your actual diffs, without waiting for external tools to catch up with the new OpenAI Responses API . The result is a minimal, focused tool you can drop into your setup today: lgaicm . It behaves like a mini aichat that does exactly one thing: generate commit messages from Git diffs, optimized for LazyGit. Why AI‑generated commit messages in LazyGit? Commit messages matter. They are the stor...
Read more

CopilotChat GlobFile Configuration

CopilotChat GlobFile Configuration Want to feed multiple files into GitHub Copilot Chat from Neovim without listing each one manually? Let's add a tiny feature that does exactly that: a file glob that includes full file contents . In this post, we'll walk through what CopilotChat.nvim offers out of the box, why the missing piece matters, and how to implement a custom #file_glob:<pattern> function to include the contents of all files matching a glob. Using Copilot Chat with Neovim CopilotChat.nvim brings GitHub Copilot's chat right into your editing flow. No context switching, no browser hopping — just type your prompt in a Neovim buffer and let the AI help you refactor code, write tests, or explain tricky functions. You can open the chat (for example) with a command like :CopilotChat , then provide extra context using built-in functions. That “extra context” is where the magic really happens. Built-in functio...
Read more