AVAComplete

A minimal VS Code extension that provides editor inline autocomplete using any user-configured HTTP endpoint. No backend, no gateway, no login, no chat — just prefix/suffix completions.

Quick start

1. Install dependencies:

bun install

1. Build the extension:

bun run build

1. Launch VS Code with the extension in development mode:

/Applications/Visual\ Studio\ Code.app/Contents/Resources/app/bin/code \
  --extensionDevelopmentPath=/Path/To/AVAComplete \
  /path/to/your/workspace

1. Open VS Code settings and configure your endpoint (see Configuration).

Build scripts

Configuration

All settings live under the avacomplete namespace.

Required settings

Optional settings

Applying settings

You can configure the extension in two ways:

Via settings.json (recommended for copy-pasting the examples below):

1. Open the Command Palette (Cmd+Shift+P / Ctrl+Shift+P)
2. Run Preferences: Open User Settings (JSON)
3. Paste the example JSON block inside the top-level object

Via the Settings UI:

1. Open Settings (Cmd+, / Ctrl+,)
2. Search for avacomplete
3. Fill in the fields under AVAComplete > Endpoint

Request template

avacomplete.endpoint.requestTemplate is a JSON object. The following placeholders are replaced before the request is sent:

Default template:

{
  "prompt": "{context}{prefix}{suffix}",
  "model": "{model}",
  "temperature": "{temperature}",
  "max_tokens": "{maxTokens}"
}

The extension parses the template as JSON, replaces placeholders, then serializes the result. Numeric placeholders must be written as string values ("{temperature}") so the JSON is valid; they are converted back to numbers before sending.

Extra context

You can opt-in to sending the current selection, clipboard, and recent edits as extra context. This is useful when the backend expects the extra information to live inside the prefix section (e.g., the AVACode [PREFIX]/[SUFFIX] format).

Context settings

Default prefixTemplate:

### Recent context
- selection: {selection}
- clipboard: {clipboard}
- recent edits: {recentEdits}

Place the {context} placeholder where you want it inside your requestTemplate. For the AVACode backend, put it inside the [PREFIX] section:

{
  "avacomplete.context.enabled": true,
  "avacomplete.endpoint.requestTemplate": "{\"model\": \"{model}\", \"messages\": [{\"role\": \"user\", \"content\": \"[PREFIX]{context}{prefix}[SUFFIX]{suffix}\"}], \"temperature\": \"{temperature}\", \"max_tokens\": \"{maxTokens}\"}"
}

Response path

avacomplete.endpoint.responsePath tells the extension where the completion text lives in the JSON response. Use dot notation and bracket indexes.

Example configurations

AVACode (Aidia)

{
  "avacomplete.endpoint.url": "https://ava-coding.aidia.it/v1/chat/completions",
  "avacomplete.endpoint.apiKey": "aidia-xxxxxxxxxxxx",
  "avacomplete.endpoint.model": "ava-autocomplete",
  "avacomplete.endpoint.temperature": 0.2,
  "avacomplete.endpoint.maxTokens": 1024,
  "avacomplete.endpoint.requestTemplate": "{\"model\": \"{model}\", \"messages\": [{\"role\": \"user\", \"content\": \"[SUFFIX]{suffix}[PREFIX]{prefix}\"}], \"temperature\": \"{temperature}\", \"max_tokens\": \"{maxTokens}\"}",
  "avacomplete.endpoint.responsePath": "choices[0].message.content"
}

The API key must start with aidia-. Replace the example key with your personal key.

Add {context} before prefix if you want to send extra context:

{
  "avacomplete.context.enabled": true,
  "avacomplete.endpoint.requestTemplate": "{\"model\": \"{model}\", \"messages\": [{\"role\": \"user\", \"content\": \"[SUFFIX]{suffix}[PREFIX]{context}{prefix}\"}], \"temperature\": \"{temperature}\", \"max_tokens\": \"{maxTokens}\"}"
}

OpenAI Completions

{
  "avacomplete.endpoint.url": "https://api.openai.com/v1/completions",
  "avacomplete.endpoint.apiKey": "sk-...",
  "avacomplete.endpoint.model": "gpt-3.5-turbo-instruct",
  "avacomplete.endpoint.temperature": 0.2,
  "avacomplete.endpoint.maxTokens": 128,
  "avacomplete.endpoint.requestTemplate": "{\"prompt\": \"{prefix}<|fim_middle|>{suffix}\", \"model\": \"{model}\", \"temperature\": \"{temperature}\", \"max_tokens\": \"{maxTokens}\"}",
  "avacomplete.endpoint.responsePath": "choices[0].text"
}

OpenAI Chat Completions

{
  "avacomplete.endpoint.url": "https://api.openai.com/v1/chat/completions",
  "avacomplete.endpoint.apiKey": "sk-...",
  "avacomplete.endpoint.model": "gpt-4o-mini",
  "avacomplete.endpoint.requestTemplate": "{\"model\": \"{model}\", \"messages\": [{\"role\": \"system\", \"content\": \"Complete the code. Return only the missing code.\"}, {\"role\": \"user\", \"content\": \"Prefix:\n{prefix}\n\nSuffix:\n{suffix}\"}], \"temperature\": \"{temperature}\", \"max_tokens\": \"{maxTokens}\"}",
  "avacomplete.endpoint.responsePath": "choices[0].message.content"
}

Ollama

{
  "avacomplete.endpoint.url": "http://localhost:11434/api/generate",
  "avacomplete.endpoint.model": "codellama:7b-code",
  "avacomplete.endpoint.requestTemplate": "{\"model\": \"{model}\", \"prompt\": \"{prefix}<FIM>{suffix}\", \"stream\": false, \"options\": {\"temperature\": {temperature}}}",
  "avacomplete.endpoint.responsePath": "response"
}

Commands

Status bar indicator

A status bar item named AVA appears at the bottom right of VS Code:

Hover over the item for a tooltip, and click it to toggle or open settings.

Packaging

To create a .vsix installable extension:

bun run build
npx @vscode/vsce package

Then install the generated .vsix from the VS Code extensions panel.

Troubleshooting

• No suggestions appear: Check that endpoint.url, endpoint.model, and endpoint.apiKey are set. Open the Extension Host output channel for errors.
• Request fails with 4xx/5xx: Verify the requestTemplate is valid JSON and matches what the endpoint expects. Test the response path against a sample response.
• VS Code complains about activation: Make sure bun run build completed and dist/extension.js exists.

Scope notes

• Chat autocomplete, Agent Manager, telemetry, login flows, CLI/TUI autocomplete, and gateway routing were intentionally removed.
• Next Edit is not included in this minimal version; it can be added later as a separate provider.