VS Code + Continue with the ORC Gateway
Continue is an open-source VS Code extension for chat, editing, and agent work. Its main advantage over VS Code's built-in BYOK is that it can also send inline autocomplete to your own model, so the ORC model can suggest code as you type.
Time needed: about 5 minutes. Prerequisites: VS Code installed locally, and an ORC LLM key that passes the key test.
Step 1 · Install the extension
Open the Extensions panel (Ctrl+Shift+X on Windows/Linux, Cmd+Shift+X on macOS), search for Continue, and install it. A Continue icon appears in the activity bar.
Step 2 · Open the config file
Click the Continue icon, then the gear icon in its panel, to open your configuration. You can also open the file directly:
| OS | Path |
|---|---|
| macOS / Linux | ~/.continue/config.yaml |
| Windows | %USERPROFILE%\.continue\config.yaml |
Step 3 · Add the ORC gateway
Replace the contents of config.yaml with the block below (or merge the models: entry into your existing file), put in your key, and save:
name: ORC LLM
version: 1.0.0
schema: v1
models:
- name: Qwen 3.8 27B (ORC)
provider: openai
model: qwen-3.8-27B
apiBase: http://129.174.189.202
apiKey: <your-virtual-key>
roles:
- chat
- edit
- apply
- autocomplete
provider: openai is not a mistake
The gateway exposes an OpenAI-compatible API, so Continue talks to it with its OpenAI client. No request goes to OpenAI; the model behind the gateway is ORC's.
The roles list controls what Continue uses the model for. Remove autocomplete if you want the model for chat only.
Step 4 · Select the model and test
Choose Qwen 3.8 27B (ORC) in the model dropdown at the bottom of the Continue chat panel, and send a short test message. Confirm you get a reply before doing real work.
To test autocomplete, open a code file and start typing a function; a grey suggestion should appear after a short pause. Press Tab to accept it.
Autocomplete speed
Autocomplete fires on nearly every keystroke pause, and a 27B chat model is slower at it than small models built for completion. If suggestions feel sluggish, remove autocomplete from roles and keep Continue for chat, or use a smaller model on your own llama.cpp server.
Keep the key out of Git
~/.continue/config.yaml lives in your home folder, not in a project, so it won't be committed by accident. If you ever create a project-level Continue config inside a repository, add it to .gitignore before saving a key in it.
Troubleshooting
| Problem | Fix |
|---|---|
| The model doesn't appear in the dropdown | The YAML didn't parse. Check indentation (spaces, not tabs) and that the file saved. Continue shows config errors at the top of its panel. |
| 404 or "not found" | Try apiBase: http://129.174.189.202/v1. Also confirm the model name with the key test. |
| 401 / authentication error | Check the apiKey value for stray spaces or quotes. |
| No autocomplete suggestions | Make sure autocomplete is listed under roles, and that autocomplete isn't switched off in Continue's settings. |
| Timeouts | Connect to the Mason VPN if you are off campus. |