Choosing Your Models

Seylo allows you to configure different LLM models for each agent to balance performance and cost. Here are recommended configurations:

Better Performance

• Planner & Validator: Claude 3.7 Sonnet

  • Better reasoning and planning capabilities

  • More reliable task validation

• Navigator: Claude 3.5 Haiku

  • Efficient for web navigation tasks

  • Good balance of performance and cost

Cost-Effective Configuration

• Planner & Validator: Claude Haiku or GPT-4o

  • Reasonable performance at lower cost

  • May require more iterations for complex tasks

• Navigator: Gemini 2.0 Flash or GPT-4o-mini

  • Lightweight and cost-efficient

  • Suitable for basic navigation tasks

Local Models

• Setup Options:

  • Use Ollama or other custom OpenAI-compatible providers to run models locally

  • Zero API costs and complete privacy with no data leaving your machine

• Recommended Models:

  • Qwen 2.5 Coder 14B

  • Mistral Small 24B

• Prompt Engineering:

  • Local models require more specific and cleaner prompts

  • Avoid high-level, ambiguous commands

  • Break complex tasks into clear, detailed steps

  • Provide explicit context and constraints

Note: The cost-effective configuration may produce less stable outputs and require more iterations for complex tasks.