Basic Agent Chat Loop

A feature-rich, interactive CLI for AWS Strands agents with token tracking, prompt templates, agent aliases, and extensive configuration options.

Features

• 🏷️ Agent Aliases - Save agents as short names (chat_loop pete instead of full paths)
• 📦 Auto-Setup - Automatically install agent dependencies from requirements.txt or pyproject.toml
• 🔔 Audio Notifications - Play sound when agent completes a turn (cross-platform support)
• 🎵 Harmony Support - Specialized processing for OpenAI Harmony format (gpt-oss models)
• 📜 Command History - Navigate previous queries with ↑↓ arrows (persisted to ~/.chat_history)
• ✍️ Multi-line Input - Type \\ to enter multi-line mode with Ctrl+D to cancel and ↑ to edit previous lines
• 💾 Session Management - Save conversations as clean markdown files in ./.chat-sessions/ (project-local)
• 📋 Copy Commands - Copy responses, queries, code blocks, or entire conversations to clipboard
• 💰 Token Tracking - Track tokens and costs per query and session
• 📝 Prompt Templates - Reusable prompts from ~/.prompts/
• ⚙️ Configuration - YAML-based config with per-agent overrides
• 📊 Status Bar - Real-time metrics (queries, tokens, duration)
• 📈 Session Summary - Full statistics displayed on exit
• 🎨 Rich Formatting - Enhanced markdown rendering with syntax highlighting
• 🔄 Error Recovery - Automatic retry logic with exponential backoff
• 🔍 Agent Metadata - Display model, tools, and capabilities

Installation

Quick Install (Recommended)

pip install basic-agent-chat-loop

That's it! The package will automatically create:

• ~/.chatrc - Configuration file with recommended defaults
• ~/.prompts/ - Sample prompt templates (on first use)

Platform-Specific Options

Windows: Command history support (pyreadline3) is now installed automatically on Windows - no extra steps needed!

AWS Bedrock integration:

pip install basic-agent-chat-loop[bedrock]

From Source

For development or the latest features:

git clone https://github.com/Open-Agent-Tools/Basic-Agent-Chat-Loop.git
cd Basic-Agent-Chat-Loop
pip install -e ".[dev]"

See docs/INSTALL.md for detailed installation instructions and troubleshooting.

Quick Start

Basic Usage

# Run with agent path
chat_loop path/to/your/agent.py

# Or use an alias (after saving)
chat_loop myagent

Agent Aliases

Save frequently used agents for quick access:

# Save an agent as an alias
chat_loop --save-alias myagent path/to/agent.py

# Use the alias from anywhere
chat_loop myagent

# List all saved aliases
chat_loop --list-aliases

# Remove an alias
chat_loop --remove-alias myagent

Example with real agents:

# Save your agents
chat_loop --save-alias pete ~/agents/product_manager/agent.py
chat_loop --save-alias dev ~/agents/senior_developer/agent.py

# Use them from anywhere
cd ~/projects/my-app
chat_loop dev  # Get coding help
chat_loop pete  # Get product feedback

Aliases are stored in ~/.chat_aliases and work from any directory.

Auto-Setup Dependencies

Automatically install agent dependencies with the --auto-setup flag (or -a for short):

# Auto-install dependencies when running an agent
chat_loop myagent --auto-setup
chat_loop path/to/agent.py -a

# Works with any of these dependency files:
# - requirements.txt (most common)
# - pyproject.toml (modern Python projects)
# - setup.py (legacy projects)

Smart detection: If you run an agent without --auto-setup and dependency files are detected, you'll see a helpful suggestion:

chat_loop myagent
💡 Found requirements.txt in agent directory. Run with --auto-setup (or -a) to install dependencies automatically

What gets installed:

• requirements.txt → pip install -r requirements.txt
• pyproject.toml → pip install -e <agent_directory>
• setup.py → pip install -e <agent_directory>

This makes sharing agents easier—just include a requirements.txt with your agent and users can install everything with one command.

Prompt Templates

The package automatically creates sample templates in ~/.prompts/ on first use:

• explain.md - Explain code in detail
• review.md - Code review with best practices
• debug.md - Help debugging issues
• optimize.md - Performance optimization suggestions
• test.md - Generate test cases
• document.md - Add documentation

Use templates in chat:

chat_loop myagent
You: /review src/app.py
You: /explain utils.py
You: /test my_function

Create custom templates:

# Create your own template
cat > ~/.prompts/security.md <<'EOF'
# Security Review

Please review this code for security vulnerabilities:

{input}

Focus on:
- Input validation
- Authentication/authorization
- Data sanitization
- Common security patterns
EOF

# Use it in chat
You: /security auth.py

Configuration

A configuration file (~/.chatrc) is automatically created on first use with recommended defaults. You can customize it to your preferences:

features:
  auto_save: true             # Automatically save conversations on exit
  show_tokens: true           # Display token counts
  show_metadata: true         # Show agent model/tools info
  rich_enabled: true          # Enhanced formatting

ui:
  show_status_bar: true       # Top status bar
  show_duration: true         # Query duration

audio:
  enabled: true               # Play sound when agent completes
  notification_sound: null    # Custom WAV file (null = bundled sound)

harmony:
  enabled: auto               # Harmony processing (auto/yes/no)
  show_detailed_thinking: true  # Show reasoning with labeled prefixes

behavior:
  max_retries: 3              # Retry attempts on failure
  timeout: 120.0              # Request timeout (seconds)

# Per-agent overrides
agents:
  'Product Pete':
    features:
      show_tokens: false
    audio:
      enabled: false          # Disable audio for this agent

Audio Notifications

Audio notifications alert you when the agent completes a response. Enabled by default with a bundled notification sound.

Platforms supported:

• macOS (using afplay)
• Linux (using aplay or paplay)
• Windows (using winsound)

Configure audio in ~/.chatrc:

audio:
  enabled: true
  notification_sound: null    # Use bundled sound

  # Or specify a custom WAV file:
  # notification_sound: /path/to/custom.wav

Per-agent overrides:

agents:
  'Silent Agent':
    audio:
      enabled: false  # Disable audio for this agent

See CONFIG.md for full configuration options.

Commands

Command	Description
#help	Show help message
#info	Show agent details (model, tools)
#context	Show token usage and context statistics
#templates	List available prompt templates
#sessions	List all saved conversation sessions
/name	Use prompt template from ~/.prompts/name.md
#resume <#>	Resume a previous session by number or ID
#compact	Save session and continue in new session with summary
#copy	Copy last response to clipboard (see variants below)
#clear	Clear screen and reset agent session
#exit, #quit	Exit chat (shows session summary)

Session Management

Save conversations automatically:

# Enable auto-save in config
features:
  auto_save: true

Resume a previous conversation:

# In chat - list sessions
You: #sessions

Available Sessions (3):
  1. MyAgent - Jan 26, 14:30 - 15 queries
     "Can you help me build a REST API..."

  2. MyAgent - Jan 25, 09:15 - 7 queries
     "Explain async/await in Python..."

# Resume by number or session ID
You: #resume 1

📋 Loading session...
✓ Found: MyAgent - Jan 26, 14:30 (15 queries, 12.5K tokens)
🔄 Restoring context...

MyAgent: I've reviewed our previous conversation about building a REST API.
We discussed Flask routing and database models. Ready to continue!

# Continue conversation with restored context
You: Let's add authentication now

Compact current session:

When your conversation gets long, use #compact to save it and start fresh while preserving context:

You: #compact

📝 Generating session summary...
💾 Saved session: myagent_20251230_143022 (15 queries, 12.5K tokens)
🔄 Starting new session with summary...

MyAgent: I've reviewed our conversation about the REST API.
We built Flask routes and database models. Ready to continue!

# Continue in new session - old queries compressed into summary
You: Now let's add authentication

View saved conversations:

Conversations are saved as clean markdown files in ./.chat-sessions/ (in current directory):

ls -lh ./.chat-sessions/
# Shows files like: simple_sally_20251230_110627.md

# View a conversation
cat ./.chat-sessions/simple_sally_20251230_110627.md

Each saved session includes an auto-generated summary that enables fast, context-aware resumption without replaying all queries.

List all saved sessions:

chat_loop --list-sessions

Sessions are saved to ./.chat-sessions/ in your current working directory, providing context separation between different projects.

Copy Commands

Quickly copy content to clipboard:

Available copy commands:

# Copy last agent response (default)
You: #copy

# Copy your last query
You: #copy query

# Copy entire conversation as markdown
You: #copy all

# Copy only code blocks from last response
You: #copy code

Example workflow:

You: Write a Python function to reverse a string

Agent: Here's a function to reverse a string:

    def reverse_string(s):
        return s[::-1]

You: #copy code
✓ Copied code blocks from last response to clipboard

# Now paste into your editor with Cmd+V (Mac) or Ctrl+V (Windows/Linux)

Multi-line Input

Press \\ to enter multi-line mode:

You: \\
... def factorial(n):
...     if n <= 1:
...         return 1
...     return n * factorial(n - 1)
...
[Press Enter on empty line to submit]

Token Tracking

During Chat

When show_tokens: true in config:

------------------------------------------------------------
Time: 6.3s │ 1 cycle │ Tokens: 4.6K (in: 4.4K, out: 237) │ Cost: $0.017

Session Summary

Always shown on exit:

============================================================
Session Summary
------------------------------------------------------------
  Duration: 12m 34s
  Queries: 15
  Tokens: 67.8K (in: 45.2K, out: 22.6K)
  Total Cost: $0.475
============================================================

Programmatic Usage

from basic_agent_chat_loop import ChatLoop

# Create chat interface
chat = ChatLoop(
    agent=your_agent,
    name="My Agent",
    description="Agent description",
    config_path=Path("~/.chatrc")  # Optional
)

# Run interactive loop
chat.run()

AWS Strands Agent Framework

This chat loop is specifically designed for AWS Strands agents with full support for:

• Automatic metadata extraction
• Tool discovery
• Streaming responses
• Token tracking and cost estimation

OpenAI Harmony Format

The chat loop includes built-in support for the OpenAI Harmony response format (designed for gpt-oss open-weight models). Harmony support is included by default in all installations.

What is Harmony?

Harmony is OpenAI's response formatting standard for their open-weight model series (gpt-oss). It provides:

• Structured conversation handling with multiple output channels
• Reasoning output generation (internal analysis separate from final response)
• Function call management with namespaces
• Tool usage tracking and structured outputs

Automatic Detection

The chat loop automatically detects Harmony agents by checking for:

• Explicit uses_harmony attribute on the agent
• Model names containing "gpt-oss" or "harmony"
• Harmony-specific methods or attributes
• Agent class names containing "harmony"

Enhanced Display

When a Harmony agent is detected, responses are automatically processed to:

• Extract and display multiple output channels (analysis, commentary, final)
• Highlight internal reasoning separately from the final response
• Detect and format tool calls appropriately
• Parse structured Harmony response formats

Configuration

Control Harmony processing behavior:

# In ~/.chatrc or .chatrc
harmony:
  enabled: auto                 # auto (default) / yes / no
  show_detailed_thinking: true  # Default - show all channels with labels

harmony.enabled options:

• auto (default) - Automatically detect harmony agents
• yes - Force enable harmony processing for all agents
• no - Disable harmony processing completely

Detailed Thinking Mode

By default, detailed thinking is enabled - showing all channels with labeled prefixes:

With detailed thinking enabled (true, default):

💭 [REASONING]
I need to analyze this query for potential bottlenecks...

📊 [ANALYSIS]
Looking at the query structure:
- Multiple table joins without proper indexes
- WHERE clause filtering happens after the joins

📝 [COMMENTARY]
This is a common pattern I see in legacy codebases...

💬 [RESPONSE]
Here are three optimizations for your database query...

To disable detailed thinking (set to false):

harmony:
  show_detailed_thinking: false  # Only show final response

Output with detailed thinking disabled:

Here are three optimizations for your database query...

Example

# Your agent using Harmony
class MyHarmonyAgent:
    uses_harmony = True  # Explicit marker

    def __call__(self, query):
        # Agent returns Harmony-formatted response
        return harmony_response

# Chat loop will automatically detect and handle Harmony format
chat_loop my_harmony_agent

Requirements

Core Dependencies

• Python 3.9+ (required by openai-harmony dependency)
• pyyaml>=6.0.1 - Configuration file parsing
• rich>=13.7.0 - Enhanced terminal rendering
• pyperclip>=1.8.0 - Clipboard support for copy commands
• openai-harmony>=0.0.8 - OpenAI Harmony format support (built-in)
• pyreadline3>=3.4.1 - Command history on Windows (auto-installed on Windows)

Optional Dependencies

• anthropic-bedrock>=0.8.0 - AWS Bedrock integration (install with [bedrock])

Built-in Features

• readline (built-in on Unix) - Command history on macOS/Linux

Platform Support

• ✅ macOS - Full support with native readline
• ✅ Linux - Full support with native readline
• ✅ Windows - Full support with automatic pyreadline3 installation

Architecture

src/basic_agent_chat_loop/
├── chat_loop.py          # Main orchestration
├── chat_config.py        # Configuration management
├── cli.py                # CLI entry point
├── components/           # Modular components
│   ├── ui_components.py      # Colors, StatusBar
│   ├── token_tracker.py      # Token/cost tracking
│   ├── template_manager.py   # Prompt templates
│   ├── display_manager.py    # Display formatting
│   ├── agent_loader.py       # Agent loading
│   └── alias_manager.py      # Alias management
docs/
├── ALIASES.md            # Alias system guide
├── CONFIG.md             # Configuration reference
├── INSTALL.md            # Installation instructions
└── Chat_TODO.md          # Roadmap and future features

Documentation

• docs/ALIASES.md - Agent alias system guide
• docs/CONFIG.md - Configuration reference
• docs/INSTALL.md - Installation instructions
• docs/Chat_TODO.md - Roadmap and future features

Development

Running Tests

# Install dev dependencies
pip install -e ".[dev]"

# Run tests
pytest

Code Quality

# Format code
black src/ tests/

# Lint
ruff check src/ tests/

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

MIT License - see LICENSE file for details.

Changelog

See CHANGELOG.md for detailed version history.

Latest Release: v1.3.3 (2024-12-24)

Hotfix release with default features enabled and harmony improvements:

• ✨ Default Features Enabled - All features now enabled by default for better UX
  • auto_save: true - Save conversations automatically
  • show_tokens: true - Display token counts and costs
  • show_status_bar: true - Status bar with agent, model, queries, time
  • show_detailed_thinking: true - Show harmony reasoning channels
• 🔧 Status Bar Fix - Status bar now displays correctly between messages
• 📊 Harmony Improvements - Enhanced detection logging and documentation
• 🎨 Better Defaults - Optimized out-of-the-box experience for new users

Troubleshooting

See docs/TROUBLESHOOTING.md for common issues and solutions.

Quick fixes:

• Package not found: Run pip install --upgrade basic-agent-chat-loop
• Command not found: Ensure pip's bin directory is in your PATH
• Import errors: Try reinstalling with pip install --force-reinstall basic-agent-chat-loop

Support

• 🐛 Bug Reports: GitHub Issues
• 💡 Feature Requests: GitHub Issues
• 📖 Documentation: docs/
• 💬 Discussions: GitHub Discussions