Advanced Features¶

This guide covers Perspt’s advanced features powered by the modern genai crate (v0.3.5), enabling sophisticated AI interactions, enhanced streaming capabilities, and productivity workflows.

Configuration Profiles and Multi-Provider Setup¶

GenAI-Powered Provider Management¶

With the genai crate integration, Perspt supports seamless switching between providers and models:

# Work profile with reasoning models
perspt --config ~/.config/perspt/work.json

# Creative profile with latest models
perspt --config ~/.config/perspt/creative.json

# Development profile with coding-focused models
perspt --config ~/.config/perspt/dev.json

# Research profile with large context models
perspt --config ~/.config/perspt/research.json

Example profile configurations:

Work Profile (work.json):

{
  "provider_type": "anthropic",
  "default_model": "claude-3-5-sonnet-20241022",
  "api_key": "${ANTHROPIC_API_KEY}",
  "providers": {
    "anthropic": "https://api.anthropic.com",
    "openai": "https://api.openai.com/v1"
  }
}

Creative Profile (creative.json):

{
  "provider_type": "openai",
  "default_model": "gpt-4.1",
  "api_key": "${OPENAI_API_KEY}",
  "providers": {
    "openai": "https://api.openai.com/v1",
    "xai": "https://api.x.ai/v1"
  }
}

Development Profile (dev.json):

{
  "provider_type": "openai",
  "default_model": "o1-mini",
  "api_key": "${OPENAI_API_KEY}",
  "providers": {
    "openai": "https://api.openai.com/v1",
    "groq": "https://api.groq.com/openai/v1"
  }
}

Research Profile (research.json):

{
  "provider_type": "google",
  "default_model": "gemini-1.5-pro",
  "api_key": "${GOOGLE_API_KEY}",
  "providers": {
    "google": "https://generativelanguage.googleapis.com",
    "anthropic": "https://api.anthropic.com"
  }
}

Enhanced Streaming and Real-time Features¶

GenAI Crate Streaming Capabilities¶

The genai crate provides sophisticated streaming with multiple event types:

Standard Streaming - Token-by-token streaming for immediate feedback - Smooth rendering with buffer management - Context-aware response building

Reasoning Model Streaming - ChatStreamEvent::Start: Beginning of response - ChatStreamEvent::Chunk: Regular content tokens - ChatStreamEvent::ReasoningChunk: Thinking process (o1-series) - ChatStreamEvent::End: Response completion

Advanced Streaming Features

# Example with reasoning model (o1-mini)
> Solve this complex math problem: ...

[Reasoning] Let me think through this step by step...
[Reasoning] First, I'll identify the key variables...
[Reasoning] Now I'll apply the quadratic formula...
[Streaming] Based on my analysis, the solution is...

Real-time Model Switching

Switch between models during conversations while maintaining context:

# Start with fast model for exploration
perspt --provider-type groq --model llama-3.1-8b-instant

# Switch to reasoning model for complex analysis
# (Context maintained across switch)
perspt --provider-type openai --model o1-mini

Model Validation and Discovery¶

Pre-flight model validation ensures reliable connections:

# Validate model before starting conversation
perspt --provider-type anthropic --model claude-3-5-sonnet-20241022 --list-models

# Discover available models for provider
perspt --provider-type google --list-models | grep gemini-2

Automatic Fallback Configuration

Configure automatic fallbacks for reliability:

{
  "provider_type": "openai",
  "default_model": "gpt-4o-mini",
  "fallback_providers": [
    {
      "provider_type": "anthropic",
      "model": "claude-3-5-haiku-20241022"
    },
    {
      "provider_type": "groq",
      "model": "llama-3.1-70b-versatile"
    }
  ]
}

Advanced Conversation Patterns¶

Multi-Model Collaborative Workflows¶

Leverage different models for their strengths within single sessions:

Research and Analysis Workflow

# 1. Start with fast model for initial exploration
perspt --provider-type groq --model llama-3.1-8b-instant

# 2. Switch to reasoning model for deep analysis
perspt --provider-type openai --model o1-mini

# 3. Use large context model for comprehensive review
perspt --provider-type google --model gemini-1.5-pro

Code Development Workflow

# Use reasoning model for architecture planning
> Design a microservices architecture for an e-commerce platform
[Using o1-mini for complex reasoning]

# Switch to coding-focused model for implementation
> Now implement the user authentication service
[Using claude-3-5-sonnet for code generation]

# Use fast model for quick iterations and testing
> Review this code for potential bugs
[Using llama-3.1-70b for rapid feedback]

Provider-Specific Optimizations¶

OpenAI Reasoning Models - Best for: Complex problem-solving, mathematical reasoning, logic puzzles - Features: Step-by-step thinking process, enhanced accuracy - Usage: Allow extra time for reasoning, provide complex multi-step problems

Anthropic Constitutional AI - Best for: Safety-critical applications, ethical reasoning, content moderation - Features: Built-in safety guardrails, nuanced understanding - Usage: Ideal for sensitive topics, business communications

Google Multimodal Capabilities - Best for: Document analysis, image understanding, large context processing - Features: 2M token context, multimodal input support - Usage: Large document analysis, comprehensive research

Groq Ultra-Fast Inference - Best for: Real-time chat, rapid prototyping, interactive sessions - Features: Sub-second response times, consistent performance - Usage: Brainstorming sessions, quick iterations

Local Model Privacy Features¶

Enhanced privacy with local Ollama integration:

Private Development Environment

{
  "provider_type": "ollama",
  "default_model": "llama3.2:8b",
  "privacy_mode": true,
  "data_retention": "none",
  "providers": {
    "ollama": "http://localhost:11434"
  }
}

Sensitive Data Processing

# Use local models for proprietary code review
perspt --provider-type ollama --model qwen2.5:14b

# Process confidential documents offline
perspt --provider-type ollama --model llama3.2:8b

Terminal UI Enhancements¶

Advanced Markdown Rendering - Syntax highlighting for code blocks - Proper table formatting and alignment - Mathematical equation rendering - Nested list and quote support

Streaming Visual Indicators - Real-time token streaming animations - Reasoning process visualization for o1-models - Connection status and model information - Error recovery visual feedback

Keyboard Shortcuts and Navigation - Input queuing while AI responds - Seamless scrolling through long conversations - Context-aware copy/paste operations - Quick model switching hotkeys

Domain Expert Prompts¶

Software Development:

{
  "system_prompt": "You are a senior software engineer with expertise in multiple programming languages, system design, and best practices. Provide detailed, practical advice with code examples when helpful. Focus on maintainability, performance, and security."
}

Academic Research:

{
  "system_prompt": "You are an academic research assistant with expertise in methodology, citation practices, and critical analysis. Provide well-researched, evidence-based responses with appropriate academic tone and references when possible."
}

Creative Writing:

{
  "system_prompt": "You are a creative writing mentor with expertise in storytelling, character development, and various literary forms. Help develop ideas, provide constructive feedback, and suggest techniques to improve writing craft."
}

Productivity and Session Management¶

Conversation Export and Archival

Perspt includes built-in conversation export functionality for productivity workflows:

# Save conversation with timestamped filename
> /save
💾 Conversation saved to: conversation_1735123456.txt

# Save with custom filename for organization
> /save python_debugging_session.txt
💾 Conversation saved to: python_debugging_session.txt

Export Features: - Raw text format without terminal styling - Chronological message order with timestamps - User and assistant messages (system messages excluded) - Automatic filename generation with Unix timestamps - Custom filename support for organized archives

Typical Export Format:

Perspt Conversation
==================
[2024-01-01 12:00:00] User: How do I optimize this Python function?
[2024-01-01 12:00:01] Assistant: I can help you optimize that function...

[2024-01-01 12:02:15] User: What about memory usage?
[2024-01-01 12:02:16] Assistant: For memory optimization, consider...

Workflow Integration:

# Research session workflow
perspt --provider-type openai --model o1-mini
# Conduct research conversation
# /save research_quantum_computing_2024.txt

# Code review session
perspt --provider-type anthropic --model claude-3-5-sonnet-20241022
# Review and discuss code
# /save code_review_auth_service.txt

# Learning session
perspt --provider-type google --model gemini-1.5-pro
# Educational conversation
# /save learning_session_rust_async.txt

Session Management Best Practices: - Use descriptive filenames with dates and topics - Export before switching contexts or models - Organize exported conversations in project folders - Archive important problem-solving sessions - Keep transcripts of architectural discussions

Context-Aware Prompts¶

Dynamic system prompts based on context:

{
  "system_prompt": "You are assisting with a ${PROJECT_TYPE} project. The user is working in ${LANGUAGE} and prefers ${STYLE} coding style. Adapt your responses accordingly and provide relevant examples."
}

Session Persistence¶

Save and Resume Conversations¶

Perspt can maintain conversation history across sessions:

{
  "conversation_history": {
    "enabled": true,
    "max_sessions": 10,
    "auto_save": true,
    "storage_path": "~/.config/perspt/history/"
  }
}

Session Commands¶

Manage conversation sessions:

> /save session_name        # Save current conversation
> /load session_name        # Load saved conversation
> /list sessions           # List all saved sessions
> /delete session_name     # Delete a saved session

Export Conversations¶

Export conversations in various formats:

> /export markdown conversation.md
> /export json conversation.json
> /export html conversation.html

Multi-Model Conversations¶

Model Comparison¶

Compare responses from different models:

> /compare "Explain quantum computing" gpt-4o claude-3-5-sonnet-20241022

This sends the same prompt to multiple models and displays responses side by side.

Model Switching¶

Switch models mid-conversation while maintaining context:

> We've been discussing Python optimization
AI: Yes, we covered several techniques including caching and algorithmic improvements.

> /model claude-3-5-sonnet-20241022
Model switched to claude-3-5-sonnet-20241022

> Can you continue with memory optimization techniques?
AI: Continuing our Python optimization discussion, let's explore memory optimization...

Plugin System¶

Perspt supports plugins for extended functionality:

Code Analysis Plugin¶

Analyze code quality and suggest improvements:

{
  "plugins": {
    "code_analysis": {
      "enabled": true,
      "languages": ["python", "javascript", "rust"],
      "features": ["linting", "security", "performance"]
    }
  }
}

Usage:

> /analyze-code
```python
def inefficient_function(data):
    result = []
    for item in data:
        if item > 0:
            result.append(item * 2)
    return result
```

Document Processing Plugin¶

Process and analyze documents:

{
  "plugins": {
    "document_processor": {
      "enabled": true,
      "supported_formats": ["pdf", "docx", "txt", "md"],
      "max_file_size": "10MB"
    }
  }
}

Usage:

> /process-document /path/to/document.pdf
> Summarize this document and highlight key points

Web Integration Plugin¶

Fetch and analyze web content:

{
  "plugins": {
    "web_integration": {
      "enabled": true,
      "allowed_domains": ["github.com", "stackoverflow.com", "docs.python.org"],
      "max_content_length": 50000
    }
  }
}

Usage:

> /fetch-url https://docs.python.org/3/library/asyncio.html
> Explain the key concepts from this documentation

Advanced Conversation Patterns¶

Role-Playing Scenarios¶

Set up specific roles for focused assistance:

> /role code_reviewer
AI: I'm now acting as a code reviewer. Please share your code for detailed analysis.

> /role system_architect
AI: I'm now acting as a system architect. Let's discuss your system design requirements.

Collaborative Problem Solving¶

Break down complex problems into manageable parts:

> /problem-solving mode
AI: I'm now in problem-solving mode. Let's break down your challenge systematically.

> I need to design a scalable microservices architecture
AI: Great! Let's approach this systematically:
    1. First, let's identify your core business domains
    2. Then we'll determine service boundaries
    3. Next, we'll design the communication patterns
    4. Finally, we'll address scalability and deployment

    Let's start with step 1: What are your main business domains?

Automation and Scripting¶

Command Scripting¶

Create scripts for common workflows:

development_workflow.perspt:

/model gpt-4
/role senior_developer
/context "Working on a ${PROJECT_NAME} project in ${LANGUAGE}"

Ready for development assistance!

Run with:

perspt --script development_workflow.perspt

Batch Processing¶

Process multiple queries in batch:

> /batch process_queries.txt

Where process_queries.txt contains:

Explain the benefits of microservices
---
Compare REST vs GraphQL APIs
---
Best practices for database design

Configuration Validation¶

Validate your configuration setup:

> /validate-config

This checks:

API key validity
Model availability
Configuration syntax
Plugin compatibility
Network connectivity

Performance Optimization¶

Response Caching¶

Cache responses for repeated queries:

{
  "cache": {
    "enabled": true,
    "ttl": 3600,
    "max_size": "100MB",
    "strategy": "lru"
  }
}

Parallel Processing¶

Process multiple requests simultaneously:

{
  "parallel_processing": {
    "enabled": true,
    "max_concurrent": 3,
    "timeout": 30
  }
}

Custom Integrations¶

IDE Integration¶

Integrate Perspt with your development environment:

VS Code Extension:

{
  "vscode": {
    "enabled": true,
    "keybindings": {
      "ask_perspt": "Ctrl+Shift+P",
      "explain_code": "Ctrl+Shift+E"
    }
  }
}

Vim Plugin:

" Add to .vimrc
nnoremap <leader>p :!perspt --query "<C-R><C-W>"<CR>

API Integration¶

Use Perspt programmatically:

import requests

def ask_perspt(question):
    response = requests.post('http://localhost:8080/api/chat', {
        'message': question,
        'model': 'gpt-4'
    })
    return response.json()['response']

Next Steps¶

Explore more advanced topics:

AI Providers - Deep dive into AI provider capabilities
Troubleshooting - Advanced troubleshooting techniques
Extending Perspt - Create custom plugins and extensions
API Reference - API reference for programmatic usage

Advanced Features¶

Configuration Profiles and Multi-Provider Setup¶

GenAI-Powered Provider Management¶

Enhanced Streaming and Real-time Features¶

GenAI Crate Streaming Capabilities¶

Model Validation and Discovery¶

Advanced Conversation Patterns¶

Multi-Model Collaborative Workflows¶

Provider-Specific Optimizations¶

Local Model Privacy Features¶

Terminal UI Enhancements¶

Domain Expert Prompts¶

Productivity and Session Management¶

Context-Aware Prompts¶

Session Persistence¶

Save and Resume Conversations¶

Session Commands¶

Export Conversations¶

Multi-Model Conversations¶

Model Comparison¶

Model Switching¶

Plugin System¶

Code Analysis Plugin¶

Document Processing Plugin¶

Web Integration Plugin¶

Advanced Conversation Patterns¶

Role-Playing Scenarios¶

Collaborative Problem Solving¶

Iterative Refinement¶

Automation and Scripting¶

Command Scripting¶

Batch Processing¶

Configuration Validation¶

Performance Optimization¶

Response Caching¶

Parallel Processing¶

Custom Integrations¶

IDE Integration¶

API Integration¶

Next Steps¶