Skip to main content

Introduction to AI Models

AI Models in Vectense Platform are large language models (LLMs) that provide intelligent processing capabilities for your workflows. They enable natural language understanding, content generation, and complex reasoning tasks.

What are AI Models?​

AI Models are sophisticated neural networks trained on vast amounts of text data that can:

  • Understand natural language input in context
  • Generate human-like text responses
  • Reason through complex problems and scenarios
  • Transform data between different formats
  • Extract specific information from unstructured content

How Models Work in Vectense​

Model Integration​

Models integrate seamlessly into workflows:

  1. Workflow Step: AI models are used as workflow steps
  2. Input Processing: Receive data from previous workflow steps
  3. Context Application: Use knowledge bases for additional context
  4. Intelligent Processing: Apply AI capabilities to the input
  5. Output Generation: Produce results for subsequent workflow steps

Persona-Driven AI​

Vectense uses a persona system to control AI behavior:

  • Role Definition: Specify what role the AI should play
  • Goal Setting: Define what the AI should accomplish
  • Skill Assignment: List specific capabilities the AI should use
  • Communication Style: Control how the AI expresses responses
  • Guardrails: Set boundaries for AI behavior

Model Types​

Cloud-Based Models​

Commercial AI services accessed via API:

Advantages:

  • Latest AI technology and capabilities
  • No infrastructure management required
  • Scalable to any volume
  • Regular updates and improvements

Considerations:

  • Data sent to external providers
  • Usage-based pricing (per token)
  • Internet connectivity required
  • Subject to provider terms and availability

Self-Hosted Models​

AI models running on your own infrastructure:

Advantages:

  • Complete data privacy and control
  • No external dependencies
  • Predictable costs (no per-use fees)
  • Customizable and fine-tunable

Considerations:

  • Requires infrastructure management
  • Initial setup and maintenance complexity
  • May have lower performance than latest cloud models
  • Hardware requirements for optimal performance

Supported Providers​

OpenAI​

Available Models:

  • Latest GPT models: Most capable models for complex reasoning and tasks
  • Optimized variants: Improved speed and efficiency options
  • Cost-effective options: Fast models for simpler tasks

Best For:

  • Complex analysis and reasoning
  • Code generation and debugging
  • Creative writing and content generation
  • General-purpose AI tasks

Anthropic​

Available Models:

  • Claude model family: Various capability and performance tiers
  • High-capability models: Complex reasoning and analysis
  • Efficient variants: Balanced performance options

Best For:

  • Document analysis and comprehension
  • Ethical and safe AI applications
  • Long-form content processing
  • Research and analytical tasks

Mistral AI​

Available Models:

  • Mistral model family: Range from small to large variants
  • High-performance options: Complex task handling
  • Efficient models: Cost-effective processing

Best For:

  • Multilingual text processing
  • Cost-effective AI processing
  • European data governance requirements
  • Efficient workflow automation

Ollama (Self-Hosted)​

Available Models:

  • Llama 2 & 3: Meta's open-source models
  • Mistral: Self-hosted version of Mistral models
  • CodeLlama: Specialized for code generation
  • Many other open-source models

Best For:

  • Sensitive data processing
  • Air-gapped environments
  • Cost-predictable deployment
  • Custom model fine-tuning

Model Capabilities​

Natural Language Understanding​

  • Text Comprehension: Understand complex documents and instructions
  • Intent Recognition: Identify what users want to accomplish
  • Context Awareness: Maintain understanding across long conversations
  • Language Detection: Automatically identify input language

Content Generation​

  • Text Creation: Generate articles, emails, summaries, and reports
  • Format Conversion: Transform content between different formats
  • Style Adaptation: Adjust writing style for different audiences
  • Template Filling: Complete forms and templates with appropriate content

Data Processing​

  • Information Extraction: Pull specific data from unstructured text
  • Classification: Categorize content into predefined groups
  • Summarization: Create concise summaries of long documents
  • Translation: Convert text between different languages

Reasoning and Analysis​

  • Logical Inference: Draw conclusions from available information
  • Problem Solving: Break down complex problems into steps
  • Comparison: Analyze similarities and differences
  • Recommendation: Suggest actions based on criteria

Model Selection Criteria​

Task Complexity​

Simple Tasks (Classification, basic extraction)

  • Smaller, efficient models from any provider
  • Faster response times, lower costs
  • Suitable for high-volume processing

Complex Tasks (Analysis, reasoning, creative work)

  • Larger, more capable models from providers
  • Higher quality outputs, better reasoning
  • Worth the additional cost for critical tasks

Data Sensitivity​

Public/Non-sensitive Data

  • Cloud models from external providers acceptable
  • Take advantage of latest AI capabilities
  • Simpler setup and management

Sensitive/Regulated Data

  • Self-hosted models recommended
  • Complete data privacy and control
  • Compliance with data governance requirements

Volume and Cost​

Low to Medium Volume

  • Cloud models with pay-per-use pricing
  • No infrastructure investment required
  • Predictable costs based on usage

High Volume

  • Consider self-hosted models
  • Infrastructure costs vs. per-token costs
  • Break-even analysis for cost optimization

Performance Requirements​

Real-time Applications

  • Faster, smaller models from any provider
  • Consider response time requirements
  • May need to trade quality for speed

Batch Processing

  • Can use more capable models
  • Quality more important than speed
  • Optimize for best results

Model Configuration​

Basic Settings​

Model Selection

  • Choose appropriate model for your use case
  • Consider performance vs. cost trade-offs
  • Test different models for quality comparison

API Configuration

  • Set API endpoints and authentication
  • Configure timeout and retry settings
  • Set up monitoring and alerting

Advanced Parameters​

Temperature (0.0 - 2.0)

  • Controls randomness in responses
  • Lower values: more deterministic
  • Higher values: more creative/random

Max Tokens

  • Limits response length
  • Helps control costs
  • Prevents overly long responses

Top P (0.0 - 1.0)

  • Alternative to temperature
  • Controls diversity of word choices
  • Fine-tunes response variation

Cost Considerations​

Token-Based Pricing​

Understanding Tokens

  • Roughly 0.75 words per token
  • Both input and output count toward costs
  • Different models have different per-token rates

Cost Optimization

  • Use efficient prompts
  • Choose appropriate model size
  • Consider self-hosted for high volume

Self-Hosted Costs​

Infrastructure Costs

  • Hardware purchase or rental
  • Electricity and cooling
  • Maintenance and support

Break-Even Analysis

  • Compare infrastructure costs to token costs
  • Consider growth projections
  • Factor in management overhead

Security and Privacy​

Data Handling​

Cloud Models

  • Data sent to external providers
  • Review privacy policies and terms
  • Understand data retention practices

Self-Hosted Models

  • Complete data control
  • No external data transmission
  • Suitable for regulated industries

Access Control​

  • Role-based access to models
  • API key management and rotation
  • Usage monitoring and auditing
  • Network security and encryption

Getting Started​

First Steps​

  1. Assess Requirements: Understand your AI processing needs
  2. Choose Provider: Select based on data sensitivity and requirements
  3. Configure Model: Set up authentication and basic parameters
  4. Test Functionality: Validate model performance with sample data
  5. Integrate Workflows: Use models in your automation processes

Best Practices​

  • Start with well-known models and standard settings
  • Test thoroughly with representative data
  • Monitor usage and costs regularly
  • Keep API keys secure and rotate regularly
  • Document successful configurations for reuse

Ready to configure your first model? Continue to Connect a Model for detailed setup instructions.