Skip to main content

Google Gemini

Gemini is the recommended LLM provider for SmarterAvatar, offering excellent RAG support through native File Search and competitive pricing.

Why Gemini?​

  • Native File Search - Built-in RAG without external vector databases
  • 1M token context - Process entire documents without chunking
  • Cost effective - Lower per-token costs than competitors
  • Multimodal - Supports text, images, and files

Setup​

1. Get API Key​

  1. Go to Google AI Studio
  2. Sign in with your Google account
  3. Navigate to Get API key
  4. Create a new API key or use existing

2. Configure Environment​

LLM_PROVIDER=gemini
GEMINI_API_KEY=AIzaSy...your-api-key
GEMINI_MODEL=gemini-3.0-flash

For RAG with native File Search:

GEMINI_FILE_SEARCH_STORE_NAME=your-knowledge-base

Available Models​

ModelContextBest ForCost (Input)
gemini-3.0-flash1M tokensMost use cases$0.50/1M
gemini-2.5-pro1M tokensComplex tasks$1.25/1M
gemini-2.5-flash1M tokensBudget option$0.15/1M

Gemini's File Search provides built-in RAG capabilities:

Upload Documents​

Use the admin panel or CLI to upload documents:

pnpm upload-gemini-file path/to/document.pdf

Supported formats:

  • PDF
  • TXT
  • Markdown
  • HTML
  • DOCX

Automatic Features​

  • Chunking - Documents automatically split for retrieval
  • Embeddings - Generated and indexed by Gemini
  • Citations - Responses include source references

File Management​

View and manage uploaded files:

pnpm list-gemini-files

Or via the Admin Panel → Knowledge Base section.

Configuration Options​

Full configuration reference:

# Required
GEMINI_API_KEY=AIzaSy...

# Model selection
GEMINI_MODEL=gemini-3.0-flash

# File Search (RAG)
GEMINI_FILE_SEARCH_STORE_NAME=my-kb

# API version (usually auto-detected)
GEMINI_API_VERSION=v1beta

Troubleshooting​

"API key not valid"​

  • Verify your API key at Google AI Studio
  • Check for extra spaces or characters in .env
  • Ensure the key has the correct permissions

"Model not found"​

  • Use the full model name (e.g., gemini-3.0-flash)
  • Check available models with pnpm list-gemini-models

"File Search not working"​

  • Ensure GEMINI_API_VERSION=v1beta is set
  • Wait a few minutes after uploading for indexing
  • Check file upload status in admin panel

Cost Estimation​

For a typical 3-minute session (4 interactions) with Gemini 3.0 Flash:

ComponentTokensCost
Input (system + RAG + user)~10,000$0.005
Output (responses)~1,200$0.0036
Total~11,200~$0.01

Gemini remains cost-effective compared to Claude/GPT-4 for most use cases.