Skip to content

An AI-powered search engine with a generative UI

4.4 KB
Raw Blame

Scripts

This directory contains utility scripts for testing and development.

Regenerate the models.dev (MIT) snapshot with bun scripts/generate-model-metadata.ts [--source <api.json>].

chat-cli.ts

A command-line interface for testing the chat API without a browser client. This script allows you to interact with the chat API directly, making it easier to debug server-side issues and test API functionality.

Features

  • Send messages to the chat API via command line
  • Real-time Server-Sent Events (SSE) streaming output
  • Configurable search modes (quick/adaptive) or disabled
  • Chat session continuity
  • Message regeneration support
  • Secure authentication via environment variables
  • URL validation for security (localhost only)

Setup

  1. Add authentication to .env.local:
    MORPHIC_COOKIES="your-cookie-string-here"
    

Usage

# Using npm script (recommended)
bun chat -m "Hello, how are you?"

# Direct usage
bun scripts/chat-cli.ts -m "Hello, how are you?"

# Disable search mode
bun chat -m "Tell me a joke" --no-search

# Use adaptive search mode for complex queries
bun chat -m "Research the latest AI developments" --search-mode adaptive

# Continue an existing chat
bun chat -c "chat_123" -m "Tell me more"

# Regenerate the last assistant message
bun chat -c "chat_123" -t regenerate --message-id "msg_456"

# Show help
bun chat --help

Options

  • -m, --message <text> - Message to send (default: "Hello, how are you?")
  • -u, --url <url> - API URL (default: http://localhost:3000/api/chat, localhost only)
  • -c, --chat-id <id> - Chat ID for session continuity (default: auto-generated)
  • -s, --search - Enable search mode with adaptive strategy (default)
  • --no-search - Disable search mode
  • --search-mode <type> - Search strategy: quick or adaptive
  • -t, --trigger <type> - Trigger type: submit (default) or regenerate
  • --message-id <id> - Message ID (required for regenerate trigger)
  • -h, --help - Show help message

Output Format

The script displays:

  • πŸš€ Request details
  • πŸ” Search mode status (quick/adaptive/disabled)
  • πŸ’¬ Chat ID for reference
  • Real-time AI responses with proper formatting
  • πŸ”§ Tool usage (when search mode is enabled)
  • βœ… Completion status

Search Modes

  • quick: Fast search with basic results
  • adaptive: Intelligent search strategy based on query type, with enhanced support for complex queries (default)
  • disabled: No search functionality (--no-search)

Advanced Usage

Message Regeneration

You can regenerate the last assistant message in a conversation:

# First, send a message and note the chat ID and message ID from the output
bun chat -m "Tell me about AI"

# Then regenerate the assistant's response
bun chat -c "chat_123" -t regenerate --message-id "msg_456"

# Or edit the user message and regenerate
bun chat -c "chat_123" -t regenerate --message-id "msg_456" -m "Tell me about machine learning instead"

Security Features

  • Authentication: Uses environment variables only (no file-based auth)
  • URL Validation: Only allows localhost and local network URLs
  • No Sensitive Logging: Cookies are never displayed in logs
  • Input Sanitization: Message length limited to 10,000 characters

Requirements

  • Bun runtime
  • Local development server running (bun dev)
  • Valid authentication cookies in .env.local

Troubleshooting

Authentication Errors

If you encounter "User not authenticated" errors:

  1. Ensure you're logged into Morphic in your browser
  2. Get fresh cookies from DevTools
  3. Update MORPHIC_COOKIES in .env.local
  4. Cookies expire after ~1 hour, so refresh them if needed

API Errors

If you encounter "Selected provider is not enabled" errors:

  1. Verify the selected model/provider is enabled in your local configuration
  2. Check the required provider API key is set in .env.local
  3. Retry with --search-mode quick to isolate search-specific issues

General Issues

  • Check the development server is running: bun dev
  • Verify .env.local exists and contains MORPHIC_COOKIES
  • Use DEBUG=1 prefix for verbose output
  • Ensure the API URL is accessible (default: http://localhost:3000/api/chat)

Command Examples for Testing

# Test basic functionality
bun chat -m "Hello, test message" --no-search

# Test adaptive search
bun chat -m "Complex analysis task" --search-mode adaptive

# Debug mode
DEBUG=1 bun chat -m "Debug test"