AI code assistants have a fundamental tension: autocomplete needs to be fast (sub-200ms or it breaks your flow), but complex refactoring needs to be smart (you want the best model available). Most setups force you to pick one model and live with the tradeoff. Fast model? Great autocomplete, terrible at multi-file refactors. Smart model? Amazing code generation, but autocomplete lags so badly you turn it off.
Antbase resolves this tension entirely. When you configure Continue to use Antbase, every request gets independently routed based on what it actually needs. Tab completions — which are just "finish this line" tasks — go to fast, cheap models with sub-second latency. But when you open the chat sidebar and ask "refactor this authentication module to use JWT with refresh tokens," that request gets escalated to a premium model that can handle the complexity. You get both fast autocomplete and smart refactoring from the same configuration.
The cost savings are significant too. In a typical coding session, 90% of AI calls are simple completions and short questions. Only 10% are the complex tasks that actually need a premium model. Without intelligent routing, you are paying premium prices for all of it. With Antbase behind Continue, you pay premium prices for only the 10% that matters.
Continue is an open-source AI code assistant for VS Code and JetBrains. It supports chat, autocomplete, and inline editing. By pointing it at Antbase, you get smart routing — simple completions go to fast models, complex refactors go to premium ones.
Install and Configure
Install the Continue extension from the VS Code marketplace. Then edit your config file at ~/.continue/config.json:
{
"models": [
{
"title": "Antbase Auto",
"provider": "openai",
"model": "auto",
"apiBase": "https://antbase.ai/v1",
"apiKey": "ant_your-api-key"
},
{
"title": "Antbase GPT-4o",
"provider": "openai",
"model": "gpt-4o",
"apiBase": "https://antbase.ai/v1",
"apiKey": "ant_your-api-key"
}
],
"tabAutocompleteModel": {
"title": "Antbase Autocomplete",
"provider": "openai",
"model": "auto",
"apiBase": "https://antbase.ai/v1",
"apiKey": "ant_your-api-key"
}
}Usage
- •Cmd+L (or Ctrl+L) opens the chat sidebar. Ask questions about your code — Antbase routes to the best model for the task.
- •Cmd+I opens inline editing. Select code, describe the change, and the model rewrites it in place.
- •Tab autocomplete works automatically as you type. These are simple completions, so Antbase typically routes to a fast model.
- •Use @file, @folder, or @codebase to include context in your prompts.
Tips
- •Add multiple model entries to switch between them in the chat dropdown — useful for comparing responses.
- •For autocomplete, fast models matter more than smart ones. Antbase handles this automatically with the "auto" model.
- •If you experience latency on autocomplete, set a dedicated fast model (e.g., model: "llama-3.3-70b") for tabAutocompleteModel.
