llms-sdk-chat

Interactive chat application demonstrating llms-sdk and llms-sdk-terminal integration with auto-generated CLI options.

Features

• Interactive Terminal UI: Built with llms-sdk-terminal differential rendering for smooth performance
• Multi-provider Support: Anthropic Claude, OpenAI, and Google Gemini through unified llms-sdk interface
• Auto-generated CLI: Command-line options automatically generated from TypeScript configuration interfaces
• Persistent Defaults: Save your preferred settings and use them automatically
• Image Support: Analyze images with multimodal models
• Streaming & Thinking: Real-time streaming with visual thinking output for supported models
• Context Management: Persistent conversation context across the session

Quick Start

# Set API keys
export ANTHROPIC_API_KEY="sk-..."
export OPENAI_API_KEY="sk-..."
export GOOGLE_API_KEY="..."

# Start chatting (no installation required)
npx @llms-sdk/chat

# Or set your defaults first
npx @llms-sdk/chat defaults anthropic -m claude-3-5-sonnet-20241022
npx @llms-sdk/chat

Commands

Interactive Chat Mode

Start an interactive chat session with persistent context:

# Using saved defaults
npx @llms-sdk/chat

# With specific provider and model
npx @llms-sdk/chat -p anthropic -m claude-3-5-sonnet-20241022
npx @llms-sdk/chat -p openai -m gpt-4o
npx @llms-sdk/chat -p google -m gemini-1.5-pro

Features:

• Persistent conversation context
• Real-time streaming output
• Visual thinking display (gray text) for supported models
• Token usage and cost tracking
• Professional terminal interface with proper input handling

Defaults System

Save and manage your preferred settings:

# Set defaults (saves to ~/.llms-sdk-chat/defaults.json)
npx @llms-sdk/chat defaults anthropic -m claude-3-5-sonnet-20241022 --thinkingEnabled
npx @llms-sdk/chat defaults openai -m gpt-4o --reasoningEffort medium

# Show current defaults
npx @llms-sdk/chat defaults --show

# Clear saved defaults
npx @llms-sdk/chat defaults --clear

# Use defaults automatically
npx @llms-sdk/chat                         # Interactive mode with defaults
npx @llms-sdk/chat "Hello world"           # One-shot with defaults

Model Discovery

Explore available models with filtering:

# List all models
npx @llms-sdk/chat models

# Filter by provider
npx @llms-sdk/chat models --provider anthropic
npx @llms-sdk/chat models --provider openai

# Filter by capabilities
npx @llms-sdk/chat models --tools               # Function calling support
npx @llms-sdk/chat models --images              # Image input support
npx @llms-sdk/chat models --cheap               # Cost-effective options

# JSON output for scripting
npx @llms-sdk/chat models --json

One-Shot Messages

Send single messages without persistent context:

# With specific provider/model
npx @llms-sdk/chat anthropic -m claude-3-5-sonnet-20241022 "Explain quantum computing"
npx @llms-sdk/chat openai -m gpt-4o "Write a poem about TypeScript"

# Using saved defaults
npx @llms-sdk/chat "What's the capital of France?"

Image Analysis

Analyze images with multimodal models:

# Single image
npx @llms-sdk/chat -i screenshot.png "What's in this image?"
npx @llms-sdk/chat anthropic -m claude-3-5-sonnet-20241022 -i diagram.jpg "Explain this flowchart"

# Multiple images
npx @llms-sdk/chat --images before.jpg after.jpg "What changed between these images?"

Supported formats: .jpg, .jpeg, .png, .gif, .webp, .bmp, .tiff, .tif

Advanced Features

Thinking/Reasoning Output

Models that support internal reasoning display their thought process:

# Anthropic thinking mode
npx @llms-sdk/chat anthropic -m claude-3-5-sonnet-20241022 --thinkingEnabled "Solve this complex problem"

# OpenAI reasoning models (automatic)
npx @llms-sdk/chat openai -m o1-mini --reasoningEffort high "Analyze this data"

Thinking appears in gray text, separated from the final response.

Provider-Specific Options

Each provider exposes its unique capabilities:

# Anthropic options
npx @llms-sdk/chat anthropic --help    # Shows thinkingEnabled, betaFeatures, etc.

# OpenAI options
npx @llms-sdk/chat openai --help       # Shows reasoningEffort, organization, etc.

# Google options
npx @llms-sdk/chat google --help       # Shows projectId, includeThoughts, etc.

Architecture

This application demonstrates several key concepts:

Auto-Generated CLI

• TypeScript interfaces automatically become command-line options
• Type validation, help text, and enum choices generated from code
• Perfect sync between TypeScript types and CLI arguments

TUI Integration

• Uses llms-sdk-terminal's differential rendering for smooth performance
• Container-based component architecture
• Proper focus management and keyboard handling

Provider Abstraction

• Single unified interface across all LLM providers
• Context serialization and conversation management
• Streaming support with provider-specific optimizations

Development

# Development mode
npm run dev

# Build
npm run build

# Type checking
npm run typecheck

# Test with simulation
npx tsx --no-deprecation src/index.ts chat --simulate-input "Hello world" "ENTER"

Testing Commands: Test model switching: npx tsx --no-deprecation src/index.ts chat --simulate-input "/" "model" "SPACE" "gpt-4o" "ENTER"

Special input keywords: "TAB", "ENTER", "SPACE", "ESC"

Examples

Quick Setup Workflow

# 1. Set your preferred provider and model
npx @llms-sdk/chat defaults anthropic -m claude-3-5-sonnet-20241022

# 2. Start using without repetitive options
npx @llms-sdk/chat "Hello!"
npx @llms-sdk/chat
npx @llms-sdk/chat -i image.png "Describe this"

Model Discovery Workflow

# Find cost-effective models with image support
npx @llms-sdk/chat models --images --cheap

# Get detailed Anthropic model information
npx @llms-sdk/chat models --provider anthropic

# Export model data for scripts
npx @llms-sdk/chat models --json > models.json

Advanced Chat Session

# Start with thinking enabled and custom limits
npx @llms-sdk/chat -p anthropic -m claude-3-5-sonnet-20241022 --thinkingEnabled --maxOutputTokens 2000

# The session will show:
# - Gray thinking text when the model reasons internally
# - Token usage per message
# - Running conversation cost
# - Smooth terminal interface with proper line handling

This application serves as both a practical chat tool and a demonstration of building sophisticated CLI applications with TypeScript, unified LLM interfaces, and modern terminal UIs.