Gemini

Gemini is Google's multimodal AI, capable of understanding code, images, and complex technical contexts. It offers strong reasoning capabilities and integration with Google's ecosystem.

Overview

Authentication

Gemini currently only supports API key authentication. Subscription-based auth (Google One AI Premium) requires browser interaction and isn't supported headlessly.

API Key Setup

1. Go to Google AI Studio
2. Click "Create API Key" or copy an existing key
3. Add to Sesame:
   • Per-user: Settings → API Keys → Gemini
   • System-wide: Set GEMINI_API_KEY environment variable

Environment Variables

Capabilities

Gemini can:

• Understand and generate code in multiple languages
• Process images and diagrams (multimodal)
• Handle very large contexts (1M+ tokens)
• Reason through complex technical problems
• Generate documentation with examples

Best Practices

Large Context

Gemini excels with large contexts:

"Analyze the entire src/ directory and suggest architectural
 improvements. Consider the current patterns and propose
 a migration path."

Multimodal Inputs

While Sesame primarily uses text, Gemini's multimodal capabilities help with:

• Understanding ASCII diagrams in comments
• Processing code that references visual concepts
• Generating documentation with diagram descriptions

Technical Documentation

Gemini is strong at documentation:

"Generate comprehensive API documentation for all public
 functions in the utils/ directory, including examples
 and edge cases."

Troubleshooting

"API key invalid"

• Verify the key at Google AI Studio
• Ensure the key hasn't been revoked
• Check for billing issues in Google Cloud Console

"Quota exceeded"

• Free tier has limited requests per minute/day
• Check your quota at Google Cloud Console
• Consider upgrading to a paid plan

"Region not supported"

• Gemini API availability varies by region
• Check supported regions
• Use a VPN if your region isn't supported (check ToS)

Slow responses

• Gemini with large contexts can be slower
• Consider using smaller context for simple tasks
• Use Gemini Flash for faster responses

CLI Reference

# Basic usage
gemini "Your prompt here"

# Specify model
gemini --model gemini-2.0-flash "prompt"

# With system instruction
gemini --system "You are a senior engineer" "review this code"

Available Models

Pricing

Free tier includes generous limits for experimentation. For current pricing, see Google AI Pricing.

Why No Subscription Auth?

Google One AI Premium includes Gemini access, but:

• Requires browser-based OAuth flow
• Session tokens expire frequently
• No official headless authentication method

For now, API keys are the only supported method. This may change if Google adds device auth support.