Autocomplete pipeline silently fails when default config.ts exists alongside config.yaml on Windows

[torresbig]

Before submitting your bug report

• I've tried using the "Ask AI" feature on the Continue docs site to see if the docs have an answer
• I'm not able to find a related conversation on GitHub discussions that reports the same bug
• I'm not able to find an open issue that reports the same bug
• I've seen the troubleshooting guide on the Continue Docs

Relevant environment info

- OS: Windows 11
- Continue version: [z.B. v0.8.50]
- IDE version: VS Code [z.B. 1.98]
- Model: qwen2.5-coder:1.5b (via Ollama)
- config:

  # Continue Configuration (2026 Optimized)
  name: Senior Java Cloud Config
  version: 1.0.0
  schema: v1

  models:
    - name: "DeepSeek V4 Flash Free (OpenRouter)"
      provider: openrouter
      model: "deepseek/deepseek-v4-flash:free"
      apiKey: "YOUR_OPENROUTER_API_KEY"
      apiBase: "[https://openrouter.ai/api/v1](https://openrouter.ai/api/v1)"
      requestOptions:
        contextLength: 16384
      systemPrompt: |
        Du bist ein erfahrener Senior Java, JavaScript und HTML Entwickler.
        Kontext: Java 18/21, HTML5, CSS3, JavaScript ES2021.
        Regeln: Antworte präzise, strikter Clean Code, SOLID.
        Nutze Records, Sealed Classes, Switch-Pattern-Matching.
        Javadoc kurz und prägnant direkt im Code.
        Code wenn möglich direkt ins projekt einfügen und ändern.

    - name: "qwen3 coder free (OpenRouter)"
      provider: openrouter
      model: "qwen/qwen3-coder:free"
      apiKey: "YOUR_OPENROUTER_API_KEY"
      apiBase: "[https://openrouter.ai/api/v1](https://openrouter.ai/api/v1)"
      requestOptions:
        contextLength: 16384
      systemPrompt: |
        Du bist ein erfahrener Senior Java, JavaScript und HTML Entwickler.
        Kontext: Java 18/21, HTML5, CSS3, JavaScript ES2021.
        Regeln: Antworte präzise, strikter Clean Code, SOLID.
        Nutze Records, Sealed Classes, Switch-Pattern-Matching.
        Javadoc kurz und prägnant direkt im Code.
        Code wenn möglich direkt ins projekt einfügen und ändern.

    - name: "DeepSeek v4 flash (OpenRouter)"
      provider: openrouter
      model: "deepseek/deepseek-v4-flash"
      apiKey: "YOUR_OPENROUTER_API_KEY"
      apiBase: "[https://openrouter.ai/api/v1](https://openrouter.ai/api/v1)"
      requestOptions:
        contextLength: 16384
      systemPrompt: |
        Du bist ein erfahrener Senior Java, JavaScript und HTML Entwickler.
        Kontext: Java 18/21, HTML5, CSS3, JavaScript ES2021.
        Regeln: Antworte präzise, strikter Clean Code, SOLID.
        Nutze Records, Sealed Classes, Switch-Pattern-Matching.
        Javadoc kurz und prägnant direkt im Code.
        Code wenn möglich direkt ins projekt einfügen und ändern.

    - name: "Qwen 2.5 Coder (local)"
      provider: openai
      model: "qwen2.5-coder:1.5b"
      apiBase: "http://localhost:11434/v1"
      requestOptions:
        contextLength: 16384
      systemPrompt: |
        Du bist ein erfahrener Senior Java, JavaScript und HTML Entwickler.
        Kontext: Java 18/21, HTML5, CSS3, JavaScript ES2021.
        Regeln: Antworte präzise, strikter Clean Code, SOLID.
        Nutze Records, Sealed Classes, Switch-Pattern-Matching.
        Javadoc kurz und prägnant direkt im Code.
        Code wenn möglich direkt ins projekt einfügen und ändern.

  # Autocomplete (lokal, kostenlos, ultraschnell)
  tabAutocompleteModel:
    name: "Qwen 2.5 Coder (local)"
    provider: openai
    model: "qwen2.5-coder:1.5b"
    apiBase: "http://localhost:11434/v1"

  # Embeddings für @Codebase
  embeddingsProvider:
    provider: openrouter
    model: "openai/text-embedding-3-small"
    apiKey: "YOUR_OPENROUTER_API_KEY"

  mcpServers: []

Describe the bug

In VS Code on Windows, the Tab Autocomplete feature breaks and shows "Setup Autocomplete model" in the UI, even though a valid local model configuration exists in config.yaml.

This happens when a default/empty config.ts (which just returns config) is present in the ~/.continue/ folder alongside a fully configured config.yaml. While the Chat/Agent subsystem is tolerant enough to read the models from the .yaml file as a fallback, the Autocomplete subsystem silently fails to initialize and falls back to a blank state without showing any clear syntax or parsing errors.

The issue is resolved immediately by deleting the .ts file and forcing a clean slate using strictly a native config.json.

Steps to reproduce

1. Set up a local Ollama model (qwen2.5-coder:1.5b) in config.yaml under tabAutocompleteModel.
2. Ensure a default config.ts exists in ~/.continue/ that simply outputs:

   export function modifyConfig(config: Config): Config { return config; }

3. Reload the Continue extension in VS Code on Windows.
4. Observe: Chat models from the YAML work fine, but Autocomplete remains completely unconfigured ("Setup Autocomplete model").

Expected behavior

Continue should either provide a clear warning that config.ts is overriding other configuration files, or ensure that the Autocomplete pipeline handles configuration file fallbacks (.yaml / .json) the exact same way the Chat pipeline does.

Environment

• OS: Windows
• IDE: VS Code
• Local LLM Provider: Ollama (Qwen 2.5 Coder)