TryAI CLI

The TryAI CLI is a comprehensive command-line tool that extends the TryAI library with interactive prompt execution, template management, and database synchronization capabilities.

Features

• Zero-config setup - Works out of the box
• Interactive REPL mode - Chat with AI models directly
• Template management - Database-backed template versioning
• Batch processing - Process multiple inputs efficiently
• Cost tracking - Monitor your AI usage and costs
• Testing & benchmarking - Validate template performance
• Template Studio - Web-based template editor (coming soon)

Installation

npm install -g @catalystlabs/tryai-cli

Quick Start

# Initialize templates (creates .tryai/config.ts and database)
tryai templates init

# Run a simple prompt
tryai prompt "Explain quantum computing in simple terms"

# Start interactive mode
tryai interactive

# Start chat mode
tryai chat --model gpt-4o

# Run a template with variables
tryai run my-template --var name=John --var topic="AI"

Commands

Basic Commands

tryai prompt <text>

Execute a simple prompt with AI.

tryai prompt "Write a haiku about coding"
tryai prompt "Explain REST APIs" --model claude-3-5-sonnet --format markdown
tryai prompt "Analyze this data" --system "You are a data scientist" --max-tokens 500

Options:

• -m, --model <model> - Model to use (default: gpt-4o-mini)
• -s, --stream - Stream the response
• -f, --format <format> - Output format (json|markdown|plain)
• -o, --output <file> - Save output to file
• --max-tokens <number> - Maximum tokens to generate
• --temperature <number> - Temperature (0-2)
• --system <text> - System prompt

tryai run <template>

Run a template with variables.

tryai run code-review --var language=python --var code="def hello(): print('hi')"
tryai run email-writer --var tone=professional --var purpose="meeting request"

Options:

• -v, --var <key=value...> - Template variables
• -m, --model <model> - Model to use
• -s, --stream - Stream the response
• -f, --format <format> - Output format
• -o, --output <file> - Save output to file

Interactive Modes

tryai interactive (alias: tryai i)

Start interactive REPL mode with a rich UI.

Features:

• Real-time chat with AI
• Slash commands (/help, /clear, /model, /cost, /exit)
• Token and cost tracking
• Model switching
• Conversation history

tryai chat

Start conversation mode in terminal.

tryai chat --model gpt-4o --system "You are a helpful coding assistant"

Options:

• -m, --model <model> - Model to use
• -s, --system <text> - System prompt
• --max-tokens <number> - Maximum tokens
• --temperature <number> - Temperature

Template Management

tryai templates init

Initialize template database and configuration.

Creates:

• .tryai/config.ts - Configuration file
• .tryai/templates.db - SQLite database (or configured database)

tryai templates list (alias: tryai templates ls)

List all available templates.

tryai templates list

tryai templates show <name>

Show template details including variables and configuration.

tryai templates show code-review

tryai templates push <file>

Push a template file to the database.

tryai templates push ./templates/email-writer.toml --name email-writer

Template file format (TOML):

[metadata]
description = "Professional email writer"
author = "Your Name"
version = "1.0.0"
tags = ["email", "business", "communication"]

[config]
model = "gpt-4o"
temperature = 0.7
max_tokens = 500

[variables]
[variables.recipient]
type = "string"
description = "Email recipient name"
required = true

[variables.purpose]
type = "string"
description = "Purpose of the email"
required = true

[variables.tone]
type = "string"
description = "Email tone"
default = "professional"

prompt = """
Write a {{tone}} email to {{recipient}} about {{purpose}}.

Make it clear, concise, and appropriate for a business context.
"""

tryai templates pull <name>

Pull a template from database to file.

tryai templates pull email-writer --output email-writer.toml

tryai templates diff <name>

Show changes between template versions.

tryai templates diff email-writer --version 2

Batch Processing

tryai batch <command>

Process multiple inputs in batch.

# From JSON array file
tryai batch "Summarize: \${input}" --input prompts.json --output results.json

# From newline-delimited file
tryai batch "Translate to Spanish: \${input}" --input sentences.txt --concurrency 5

Input file formats:

• JSON array: ["text 1", "text 2", "text 3"]
• Newline-delimited: Each line is a separate input

Options:

• -i, --input <file> - Input file (required)
• -o, --output <file> - Output file
• -c, --concurrency <number> - Concurrent requests (default: 3)
• -m, --model <model> - Model to use

Testing & Benchmarking

tryai test <template>

Test a template with sample data.

tryai test email-writer --var recipient=John --var purpose="meeting" --samples 5
tryai test json-parser --validate-json --samples 10

Options:

• -v, --var <key=value...> - Template variables
• -s, --samples <number> - Number of test samples (default: 1)
• --validate-json - Validate JSON output format
• --validate-schema <file> - Validate against JSON schema

tryai benchmark <template>

Performance benchmark a template.

tryai benchmark code-review --iterations 50 --concurrency 3
tryai benchmark fast-summary --iterations 100 --warmup 5

Options:

• -n, --iterations <number> - Number of iterations (default: 10)
• -c, --concurrency <number> - Concurrent requests (default: 1)
• --warmup <number> - Warmup iterations (default: 2)

Statistics & Environment

tryai stats

Show usage statistics.

tryai stats --detailed

tryai env list

List current configuration.

tryai env list

tryai env use <provider> [model]

Set default provider and model.

tryai env use openai gpt-4o
tryai env use anthropic claude-3-5-sonnet

Configuration

Config File Structure

The CLI creates a .tryai/config.ts file for configuration:

import type { TryAIConfig } from '@catalystlabs/tryai-cli';

const config: TryAIConfig = {
  defaultModel: 'gpt-4o-mini',
  defaultProvider: 'openai',
  
  // Template database configuration
  templates: {
    database: {
      type: 'sqlite',
      connection: '.tryai/templates.db'
    }
  },
  
  // API keys (can also use environment variables)
  apiKeys: {
    openai: process.env.OPENAI_API_KEY,
    anthropic: process.env.ANTHROPIC_API_KEY,
    google: process.env.GOOGLE_API_KEY,
    llama: process.env.LLAMA_API_KEY,
  }
};

export default config;

Environment Variables

Set API keys via environment variables:

export OPENAI_API_KEY="your-key-here"
export ANTHROPIC_API_KEY="your-key-here"
export GOOGLE_API_KEY="your-key-here"
export LLAMA_API_KEY="your-key-here"

Database Support

The CLI supports multiple database types for template storage:

• SQLite (default): Local file-based database
• PostgreSQL: For team/production use
• MySQL: Alternative production option

// PostgreSQL example
templates: {
  database: {
    type: 'postgres',
    connection: 'postgresql://user:password@localhost:5432/tryai'
  }
}

// MySQL example
templates: {
  database: {
    type: 'mysql',
    connection: 'mysql://user:password@localhost:3306/tryai'
  }
}

Examples

Creating a Code Review Template

1. Create code-review.toml:

[metadata]
description = "AI-powered code review assistant"
tags = ["code", "review", "quality"]

[config]
model = "gpt-4o"
temperature = 0.3

[variables]
[variables.language]
type = "string"
description = "Programming language"
required = true

[variables.code]
type = "string"
description = "Code to review"
required = true

[variables.focus]
type = "string"
description = "Review focus (security, performance, style)"
default = "general"

prompt = """
Review this {{language}} code with focus on {{focus}}:

```{{language}}
{{code}}

Provide:

1. Overall assessment
2. Specific issues found
3. Suggestions for improvement
4. Security considerations (if applicable)

Be constructive and educational in your feedback. """

2. Push to database:

```bash
tryai templates push code-review.toml

3. Use the template:

tryai run code-review \
  --var language=python \
  --var focus=security \
  --var code="def authenticate(password): return password == 'admin'"

Batch Processing Example

Process multiple code files for review:

1. Create files.json:

[
  "def user_login(pwd): return pwd == 'password123'",
  "SELECT * FROM users WHERE id = '" + user_id + "'",
  "function validateEmail(email) { return true; }"
]

2. Run batch review:

tryai batch "Review this code for security issues: \${input}" \
  --input files.json \
  --output security-review.json \
  --model gpt-4o \
  --concurrency 3

Performance Testing

Test and benchmark your templates:

# Test with various inputs
tryai test code-review \
  --var language=javascript \
  --var code="console.log('test')" \
  --samples 10

# Benchmark performance
tryai benchmark code-review \
  --iterations 50 \
  --concurrency 3 \
  --warmup 5

Coming Soon

• Template Studio: Web-based template editor with visual workflow builder
• Team Collaboration: Share templates across team members
• Template Marketplace: Discover and share templates with the community
• Advanced Analytics: Detailed usage and performance analytics
• CI/CD Integration: Template testing in continuous integration pipelines

Support

For issues and feature requests, visit: GitHub Issues

License

MIT License - see LICENSE file for details.