Models

About our models

Yachtsy provides four specialized AI models, each optimized for different marine and sailing use cases. All models support real-time streaming and are accessible through the same OpenAI-compatible API.

Available Models

yachtsy-agent

General sailing assistant with intelligent routing

• Type: Router-based with tool selection
• Streaming: Chunked (optimized for mixed content)
• Best For: General sailing questions, conversational assistance
• Features: Automatically routes to specialized tools based on query type

response = client.chat.completions.create(
    model="yachtsy-agent",
    messages=[{"role": "user", "content": "How do I prepare for a long passage?"}]
)

yachtsy-deep-research-agent

Comprehensive research and analysis

• Type: Direct LLM with structured analysis
• Streaming: Token-by-token (real-time)
• Best For: Market studies, competitive analysis, detailed investigations
• Features: Executive summaries, market context, technical deep dives, recommendations

response = client.chat.completions.create(
    model="yachtsy-deep-research-agent",
    messages=[{"role": "user", "content": "Comprehensive analysis of Catalina 34 vs Hunter 34"}],
    stream=True
)

Response Structure:

• Executive Summary
• Detailed Analysis
• Market Context
• Technical Deep Dive
• Research Sources & Methodology
• Recommendations

yachtsy-listings-agent

Fast boat listings search

• Type: Tool-based with LLM formatting
• Streaming: Real-time formatting
• Best For: Finding boats for sale, inventory search
• Features: Searches 25,000+ listings, formatted results, price/location filtering

response = client.chat.completions.create(
    model="yachtsy-listings-agent", 
    messages=[{"role": "user", "content": "Find Beneteau 40 boats under $200k"}],
    stream=True
)

Data Sources:

• SailboatListings.com
• YachtWorld
• BoatTrader
• Regional brokers

yachtsy-survey-agent

Professional boat surveys and inspections

• Type: Tool-based with vision analysis
• Streaming: Real-time analysis
• Best For: Boat condition assessment, survey reports, inspection analysis
• Features: Image processing, condition evaluation, detailed reports

response = client.chat.completions.create(
    model="yachtsy-survey-agent",
    messages=[{"role": "user", "content": "Survey this Catalina 34 listing: https://example.com/listing"}],
    stream=True
)

Analysis Areas:

• Hull condition
• Rigging assessment
• Engine evaluation
• Electrical systems
• Interior condition

Model Comparison

Usage Examples

Research Analysis

curl -X POST https://api.yachtsy.ai/v1/chat/completions \
  -H "Authorization: Bearer YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "yachtsy-deep-research-agent",
    "messages": [{"role": "user", "content": "Market analysis of 35-40ft cruising sailboats"}],
    "stream": true
  }'

Boat Search

curl -X POST https://api.yachtsy.ai/v1/chat/completions \
  -H "Authorization: Bearer YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "yachtsy-listings-agent", 
    "messages": [{"role": "user", "content": "Catalina 34 boats in California under $50k"}],
    "stream": true
  }'

Boat Survey

curl -X POST https://api.yachtsy.ai/v1/chat/completions \
  -H "Authorization: Bearer YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "yachtsy-survey-agent",
    "messages": [{"role": "user", "content": "Analyze this boat listing for potential issues"}],
    "stream": true
  }'

Streaming Performance

All models support optimized streaming with different characteristics:

Token-by-Token Streaming (Deep Research)

• Latency: ~50ms per token
• Experience: Text appears as AI generates it
• Best For: Research analysis, detailed explanations

Real-Time Formatting (Listings/Survey)

• Latency: ~5ms per chunk
• Experience: Formatted content appears smoothly
• Best For: Structured data, reports, listings

Chunked Streaming (General Agent)

• Latency: ~10ms per chunk
• Experience: Consistent word-group delivery
• Best For: Conversational responses, mixed content

Authentication & API Keys

All models use the same authentication system:

Authorization: Bearer yk-user-{userId}-{randomBytes}

Key Features:

• Per-user attribution and tracking
• Usage analytics and billing integration
• Firestore-based storage and management
• Web interface for key creation and management

Error Handling

All models return consistent error formats:

{
  "error": {
    "message": "Invalid model specified",
    "type": "invalid_request_error", 
    "code": "model_not_found"
  }
}

Model-Specific Errors:

• Deep Research: May timeout on extremely complex queries (90s limit)
• Listings: May return empty results if no matches found
• Survey: Requires valid listing URLs or image data
• General: May route to unavailable tools in rare cases