Solution: GPT-OSS Agent Mode Workaround Using Ollama Model Aliases

[lasseleendertz]

Update (7. Aug 2025)

Continue v1.1.70+ on the VS Code Marketplace (Git tag v1.0.20-vscode) now should support GPT-OSS models natively.
PR #7006
If you’re already on that version, you can hopefully skip this alias workaround - otherwise, see below for a quick fix or just to know for future models.

Solution: GPT-OSS Agent Mode Workaround Using Ollama Model Aliases

I encountered the same issue as others with GPT-OSS models where the Agent and Plan mode was greyed out.
Here's a workaround solution that doesn't require complex configuration changes:

Root Cause

Continue has an internal whitelist of model names that are recognized as "agent-capable". GPT-OSS model names like gpt-oss:120b are not on this list, causing the Agent mode to be disabled.

Solution: Create Model Aliases in Ollama

Instead of trying to trick Continue with requestOptions or complex configurations, simply create an alias in Ollama:

# Create an alias from your GPT-OSS model to a recognized name
ollama cp gpt-oss:120b gpt-4o

# Verify the alias was created
ollama list

Continue Configuration

I used this basic configuration with the aliased name:

{
  "models": [
    {
      "title": "GPT-OSS 120B (Local)",
      "provider": "openai",
      "model": "gpt-4o",                       // the alias Continue accepts
      "apiBase": "http://localhost:11434/v1",  // or your ngrok URL (I used ngrok)
      "apiKey": "ollama",                      // default for local Ollama
      "roles": ["chat", "edit", "apply"],
      "capabilities": ["tool_use"],
      "supportsTools": true
    }
  ]
}

Result

• ✅ Agent/Plan mode is no longer greyed out
• ✅ Your original GPT-OSS model remains unchanged
• ✅ Both model names work (gpt-oss:120b and gpt-4o)

Why This Works

Continue recognizes gpt-4o as agent-capable, while Ollama serves your actual GPT-OSS model under this alias. It's a clean solution that works with Continue's current model recognition logic.

Here’s a concise “undo” note you can append to the GitHub comment so everyone knows how to roll back the workaround later:

---

How to Undo the Workaround

When Continue adds native support for GPT-OSS model names, you can revert in two quick steps:

1. Delete the alias in Ollama

# Remove only the alias tag – the original model stays untouched
ollama rm gpt-4o

2. Update your Continue config

{
  "models": [
    {
      "model": "gpt-oss:120b",   // switch back to the original name
      "apiBase": "http://localhost:11434/v1",
      "provider": "openai"
      "apiKey": "ollama"
    }
  ]
}

That’s it—no data loss, no extra cleanup needed.

Tested with: Continue extension, Ollama, GPT-OSS 120B model

Hope this helps others with the same issue! ❤️

-Lasse

[612crypto]

This worked great, thank you!