tafysh 0.1.0

AI-enhanced terminal shell with LLM-powered capabilities

AgentSH

AI-Enhanced Terminal Shell with LLM-Powered Capabilities

AgentSH wraps traditional shells (Bash/Zsh/Fish) with intelligent features:

• Natural language to command translation
• Multi-step autonomous task execution
• Multi-device orchestration (servers, robots, IoT)
• Strong security with human-in-the-loop
• Memory and context management

Installation

Requires uv for dependency management.

# Install uv (if not already installed)
curl -LsSf https://astral.sh/uv/install.sh | sh

# From PyPI (when released)
uv pip install agentsh

# From source
git clone https://github.com/TafyLabs/AgentSH.git
cd AgentSH
uv sync
uv pip install -e ".[dev]"

Quick Start

# Set your API key
export ANTHROPIC_API_KEY="your-key-here"

# Start AgentSH
uv run agentsh

# Or with custom config
uv run agentsh --config ~/.agentsh/config.yaml

Usage

Inside the AgentSH shell:

# Regular shell commands work normally
ls -la
git status

# AI-powered requests (prefix with 'ai ')
ai list all python files modified today
ai explain what this git command does: git rebase -i HEAD~3
ai help me set up a Python virtual environment

# Force shell execution (prefix with !)
!echo "force shell mode"

Configuration

Create ~/.agentsh/config.yaml:

llm:
  provider: anthropic
  model: claude-sonnet-4-20250514

shell:
  backend: zsh
  ai_prefix: "ai "

security:
  mode: normal
  require_confirmation: true

log_level: INFO

See examples/config.yaml for full options.

Development

# Install dev dependencies
make dev

# Or manually with uv
uv pip install -e ".[dev,all]"

# Run tests
make test

# Run linters
make lint

# Run type checks
make type-check

# Run all checks
make check

# Sync dependencies
make sync

# Update lock file
make lock

Architecture

AgentSH is built with a modular architecture:

agentsh/
├── shell/         # User I/O, PTY management
├── agent/         # LLM client, planning, execution
├── tools/         # Tool interface, registry
├── workflows/     # LangGraph orchestration
├── memory/        # Session & persistent storage
├── security/      # Risk classification, RBAC
├── telemetry/     # Logging, metrics, health
├── orchestrator/  # Multi-device, SSH, MCP
├── plugins/       # Extensible toolsets
└── config/        # Configuration management

See docs/ARCHITECTURE_SUMMARY.md for details.

Implementation Status

Phase	Component	Status
0	Foundation	✅ Complete
1	Shell Wrapper	✅ Complete
2	LLM Integration	✅ Complete
3	Security	✅ Complete
4	Tool Interface	✅ Complete
5	Workflows	✅ Complete
6	Memory	✅ Complete
7	Telemetry	✅ Complete
8	Orchestration	✅ Complete
9	Robotics	✅ Complete
10	Polish	✅ Complete
11	Testing & Coverage	✅ Complete
12	Release & Distribution	✅ Complete

See docs/IMPLEMENTATION_CHECKLIST.md for details.

Security

AgentSH includes multiple security layers:

• Risk Classification: Commands classified as SAFE, MEDIUM, HIGH, CRITICAL
• RBAC: Role-based access control (VIEWER, OPERATOR, ADMIN)
• Human-in-the-Loop: Approval required for dangerous operations
• Audit Logging: All actions logged for compliance

Contributing

Contributions are welcome! Please read the documentation and submit PRs.

License

Apache License 2.0 - see LICENSE for details.