AI Provider Configuration

Configure Chrome AI, OpenAI, or Ollama for VAssist’s language model

VAssist supports three AI providers for natural language processing. Choose the provider that best fits your needs.

Provider Overview

Chrome AI

Free, local, and private

Uses Google’s Gemini Nano running entirely on your device. No API keys required.

OpenAI

Powerful cloud models

Access GPT-4, GPT-4 Turbo, and other OpenAI models via API.

Ollama

Self-hosted flexibility

Run local LLMs (Llama 2, Mistral, etc.) on your own machine.

Chrome AI (Default)

Requirements

Chrome 138+ required with specific feature flags enabled.

Chrome AI uses Google’s on-device Gemini Nano model. Configuration is done via chrome://flags:

Required Flags

1. Enable Optimization Guide

   Navigate to:

   chrome://flags/#optimization-guide-on-device-model
   

   Set to: “Enabled BypassPerfRequirement”

   This enables the on-device AI model to run without hardware restrictions.

2. Enable Prompt API

   Navigate to:

   chrome://flags/#prompt-api-for-gemini-nano
   

   Set to: “Enabled”

   Activates the Language Model API for text generation.

3. Enable Multimodal Input

   Navigate to:

   chrome://flags/#multimodal-input
   

   Set to: “Enabled”

   Required for audio/image support in Chrome AI.

4. Restart Chrome

   After enabling all flags, restart Chrome for changes to take effect.

Configuration

provider · string · default: "chrome-ai"

Set provider to chrome-ai in AI Config settings.

chromeAi.temperature · number · default: "1.0"

Controls response randomness and creativity.

• Range: 0.0 - 2.0
• Low (0.0-0.7): More focused and deterministic
• Medium (0.8-1.2): Balanced creativity
• High (1.3-2.0): More random and creative

chromeAi.topK · number · default: "3"

Number of top token candidates considered for each generation step.

• Range: 1 - 128
• Lower: More focused responses
• Higher: More diverse vocabulary

chromeAi.outputLanguage · string · default: "en"

Preferred output language for responses.

Supported languages:

• en - English
• es - Spanish
• ja - Japanese

chromeAi.enableImageSupport · boolean · default: "true"

Enable multimodal image analysis capabilities.

Requires the Multimodal Input flag enabled.

chromeAi.enableAudioSupport · boolean · default: "true"

Enable multimodal audio transcription.

Requires the Multimodal Input flag enabled.

System Prompts

Customize the assistant’s personality and behavior:

Available Personalities

chromeAi.systemPromptType · string · default: "default"

Select a pre-configured personality:

• default: Helpful and concise assistant
• professional: Formal, well-structured responses
• friendly: Warm and conversational
• technical: Expert technical guidance
• creative: Imaginative and expressive
• concise: Brief, to-the-point answers
• teacher: Educational explanations
• custom: Use your own custom prompt

chromeAi.systemPrompt · string · default: ""

Custom system prompt (only used when systemPromptType is custom).

Example:

You are a helpful coding assistant specializing in JavaScript and React.
Provide clear, concise code examples and explanations.

OpenAI

API Key Setup

1. Get API Key

   1. Sign up at platform.openai.com
   2. Navigate to API Keys section
   3. Click “Create new secret key”
   4. Copy the key (starts with sk-)
2. Configure VAssist

   1. Open AI Config settings
   2. Select OpenAI as provider
   3. Paste your API key
   4. Choose a model

Configuration

provider · string

Set to openai to use OpenAI models.

openai.apiKey · string · required

Your OpenAI API key.

Never share or commit your API key. It’s stored locally in browser storage.

openai.model · string · default: "gpt-4-turbo-preview"

OpenAI model to use for chat completions.

Recommended models:

• gpt-4-turbo-preview - Latest GPT-4 Turbo (best quality)
• gpt-4 - Standard GPT-4
• gpt-3.5-turbo - Faster, more economical
• gpt-3.5-turbo-16k - Extended context window

openai.temperature · number · default: "0.7"

Response creativity (0.0 - 2.0).

• 0.0-0.3: Focused, deterministic
• 0.4-0.7: Balanced (recommended)
• 0.8-1.0: Creative
• 1.1-2.0: Very creative/random

openai.maxTokens · number · default: "2000"

Maximum tokens in the response.

• Higher values allow longer responses
• Affects API costs
• Typical range: 500-4000

openai.enableImageSupport · boolean · default: "true"

Enable GPT-4 Vision for image analysis.

Requires gpt-4-vision-preview or similar vision-capable model.

openai.enableAudioSupport · boolean · default: "true"

Enable audio transcription via Whisper API.

System Prompts

openai.systemPromptType · string · default: "default"

Choose from the same personality options as Chrome AI (default, professional, friendly, etc.).

openai.systemPrompt · string · default: ""

Custom system prompt when systemPromptType is custom.

Ollama (Local)

Setup

1. Install Ollama

   Download and install from ollama.ai

   Supported platforms: macOS, Linux, Windows

2. Pull a Model

   Open terminal and download a model:

   # Llama 2 (7B)
   ollama pull llama2
   
   # Mistral (7B)
   ollama pull mistral
   
   # CodeLlama (7B)
   ollama pull codellama
   
3. Start Ollama Server

   Ollama runs automatically after installation. Verify it’s running:

   curl http://localhost:11434/api/tags
   
4. Configure VAssist

   1. Open AI Config
   2. Select Ollama as provider
   3. Set endpoint to http://localhost:11434
   4. Enter the model name (e.g., llama2)

Configuration

provider · string

Set to ollama for local Ollama models.

ollama.endpoint · string · default: "http://localhost:11434"

Ollama API endpoint.

• Default: http://localhost:11434
• Custom: If running on different port/host

ollama.model · string · default: "llama2" · required

Model name to use.

Popular models:

• llama2 - Meta’s Llama 2
• mistral - Mistral 7B
• codellama - Code-specialized Llama
• mixtral - Mixtral 8x7B (larger, more capable)
• phi - Microsoft Phi-2

Run ollama list to see installed models.

ollama.temperature · number · default: "0.7"

Response creativity (0.0 - 2.0).

ollama.maxTokens · number · default: "2000"

Maximum tokens in response.

ollama.enableImageSupport · boolean · default: "true"

Enable image analysis (requires vision-capable model like llava).

ollama.enableAudioSupport · boolean · default: "true"

Enable audio transcription (experimental).

System Prompts

ollama.systemPromptType · string · default: "default"

Personality selection (same options as Chrome AI and OpenAI).

ollama.systemPrompt · string · default: ""

Custom system prompt for custom personality type.

Provider Comparison

Feature	Chrome AI	OpenAI	Ollama
Cost	Free	Pay per token	Free (self-hosted)
Privacy	Local, private	Cloud, data sent to OpenAI	Local, private
Setup	Chrome flags	API key	Install + download models
Performance	Fast, on-device	Fast, cloud-powered	Depends on hardware
Models	Gemini Nano	GPT-4, GPT-3.5	Llama 2, Mistral, etc.
Image Support	Yes (multimodal)	Yes (GPT-4 Vision)	Yes (with llava)
Audio Support	Yes (multimodal)	Yes (Whisper)	Limited
Internet Required	No	Yes	No

Troubleshooting

Chrome AI not available

Check:

1. Chrome version 138 or higher
2. All three flags enabled in chrome://flags
3. Chrome restarted after enabling flags
4. Model downloaded (check chrome://components)

Model download: Chrome downloads Gemini Nano automatically. This may take 10-30 minutes.

OpenAI API errors

Common issues:

• Invalid API key: Double-check key in settings
• Rate limit: Reduce request frequency or upgrade plan
• Model not found: Verify model name spelling
• Insufficient quota: Add credits to OpenAI account

Ollama connection failed

Verify:

1. Ollama is running: ollama list
2. Endpoint is correct: http://localhost:11434
3. Model exists: ollama list shows your model
4. No firewall blocking port 11434

Restart Ollama:

# macOS/Linux
ollama serve