Question 1

Why does Continue show an "Invalid URL" error when calling a model?

Accepted Answer

The usual cause is an `apiBase` value without a URL protocol. Use a full absolute URL:

```yaml
# Bad
apiBase: localhost:11434

# Good
apiBase: http://localhost:11434
```

For HTTPS or OpenAI-compatible providers, use the complete base path, for example:

```yaml
provider: openai
apiBase: https://aimoway.com/v1
apiKey: YOUR_API_KEY
model: your-model-name
```

For local Ollama, you can usually omit `apiBase` and let Continue use its default. If you set it explicitly, include `http://` or `https://`.

If the model comes from the Continue hub, check that secrets such as `${{ secrets.MY_KEY }}` resolve correctly. A literal template string left in the configuration means the secret is missing or misnamed.

A code fix also normalizes bare `apiBase` values by prepending `http://`, validates the final URL, reports the provider and invalid value, and guards against `undefined` `apiBase` in `getMaxStopWords()`. Update Continue to a release containing that fix if the configuration change alone does not resolve it.

Question 2

How do I fix Continue's 401 "Missing Authentication header" with OpenRouter when using ${env:OPENROUTER_API_KEY} over VS Code Remote-SSH?

Accepted Answer

Upgrade the Continue extension running on the Remote-SSH host to a version containing the fix for YAML environment placeholder resolution, then run "Developer: Reload Window" or restart the remote VS Code server. Keep the config as `apiKey: ${env:OPENROUTER_API_KEY}`. If the variable is only defined in an interactive `~/.bashrc`, the extension host may not inherit it even though the integrated terminal and `curl` can read it; restart the VS Code server after exporting the variable or reconnect from a login session where it is available. As a temporary workaround, hardcode the key directly in `config.yaml` with `apiKey: sk-or-v1-...`, then remove it after upgrading.

Question 3

How do I fix `.continue/rules` userRules appearing twice in the Continue CLI system prompt?

Accepted Answer

This is a bug in Continue CLI 1.5.47: rules loaded from `.continue/rules/*.md` can be merged from both the config and markdown rule sources, causing the full `userRules` block to be duplicated. Upgrade `@continuedev/cli` to the latest release containing the fix `fix(cli): dedupe userRules from config and markdown rule sources`:

```bash
npm install -g @continuedev/cli@latest
continue --version
```

As a temporary workaround, ensure the same rules are not defined in both `.continue/rules/*.md` and your config's `userRules`; keep them in only one source until the upgraded CLI is installed.

Question 4

How to fix 'Token limit reached. File/range likely too large for this edit' error in Continue edit mode with Ollama?

Accepted Answer

Edit mode (Ctrl+I) sends the ENTIRE file content plus the edit prompt template, not just your selection, so it consumes far more tokens than chat mode. With `contextLength: 8192` the limit is hit even on small files. Fix:

1. Increase `contextLength` in `config.yaml` (deepseek-r1 supports up to 128k):
```yaml
defaultCompletionOptions:
  contextLength: 32768
```
2. Critical: Ollama ignores this and defaults `num_ctx` to 4096. Override it per-model:
```yaml
requestOptions:
  extraBodyProperties:
    options:
      num_ctx: 32768
```
Or set it globally via environment variable `OLLAMA_CONTEXT_LENGTH=32768` before starting Ollama.
3. If still failing, select a smaller code range before pressing Ctrl+I, or switch the `edit` role to a model with a larger native context window.

Note: raising context length increases VRAM/RAM usage, so lower it if the 8b model becomes slow or OOMs.

continue

Overview

README Preview

FAQ (4)

同类型项目

hermes-agent

firecrawl

go

markitdown