CLIO - Command Line Intelligence Orchestrator

June 16, 2026 · View on GitHub

A terminal-native AI coding tool that reads your code, edits files, runs commands, uses git, and works through tasks with you.

I built CLIO for myself. I spend more time in terminal sessions than I do using GUIs, and I wanted a terminal-first AI development tool that worked the way I work. It didn't really exist, so I built it. Starting with version 20260119.1, CLIO has been building itself - all development on SAM, CLIO, and ALICE is done through pair programming with AI agents using CLIO.

GPL-3.0 License Perl 5.32+ Discussions

What You Can Do With CLIO

  • Give it a task, it does the work - CLIO investigates your codebase, proposes a plan, you approve, it implements. Edits files, runs tests, commits changes.
  • Work from anywhere - Local shells, SSH sessions, tmux, Docker, headless servers. Anywhere Perl runs.
  • Zero dependencies - Pure Perl with standard core modules. No CPAN, no npm, no pip. Install and run.
  • Pick up where you left off - Persistent sessions with full history. Long-term memory carries across projects.
  • Coordinate parallel agents - Spawn sub-agents with file locks, git locks, and rate limiting.
  • Run across your fleet - SSH into any machine, deploy CLIO, run a task, get results.
  • Stay private - Secret redaction catches API keys and tokens before they reach the AI. Your code stays on your machine.
  • Interrupt anytime - Press any key to stop mid-task. CLIO pauses, asks what you need, and adapts.

CLIO works like pair programming, not prompt-and-response chat. You describe the task, CLIO investigates your code, proposes a plan, and implements after your approval.

Core Features

CategoryCapabilities
FilesRead, write, search, edit, manage files
GitStatus, diff, commit, branch, push, pull, stash, tag, worktree
TerminalExecute commands and scripts directly
RemoteRun AI tasks on remote systems via SSH
Multi-AgentSpawn parallel agents with file locks, git locks, and rate limiting
MultiplexerLive agent output panes via tmux, GNU Screen, or Zellij
MemoryShort-term sessions plus long-term memory across projects
ProfileLearns your working style and personalizes collaboration
TodosManage tasks within your workflow
UndoRevert AI changes from any turn
SkillsCustom AI skills loaded from Git repositories
PluginsDrop-in tool extensions
OpenSpecSpec-driven development with AI-generated artifacts
WebFetch and analyze web content
MCPConnect to external tool servers via Model Context Protocol

Slash Commands

Type /help in any session for the full list. Highlights:

  • /api - providers, models, login, thinking
  • /config - global settings, persistence
  • /session - history, switch, export
  • /memory - long-term patterns
  • /skills - manage custom skills
  • /spec - OpenSpec lifecycle
  • /agent - sub-agents (spawn, inbox, status)
  • /mcp - Model Context Protocol servers
  • /undo - revert changes from the last turn
  • /profile - working-style profile
  • /update - in-place version updates
  • /usage - billing and token usage
  • /stats - performance and memory

See docs/USER_GUIDE.md for the complete reference.

AI Providers

CLIO ships with 15 provider configurations. All API providers are OpenAI-compatible; Anthropic and Google have native protocol adapters.

ProviderAuthBest For
GitHub CopilotOAuthMultiple models (GPT, Claude, MiniMax), easiest setup
AnthropicAPI KeyClaude models, extended thinking, native API
OpenAIAPI KeyGPT and o-series models
Google GeminiAPI KeyLarge context (1M+), multimodal, native API
DeepSeekAPI KeyCoding, reasoning
OpenRouterAPI KeyAccess to hundreds of models
Ollama CloudAPI KeyHosted open-source models (Qwen, Gemma, DeepSeek)
MiniMaxAPI KeyHigh-throughput coding, 1M context
Z.AIAPI KeyGLM-5 models, long-horizon tasks, vision
NVIDIA NIMAPI KeyNemotron, Llama, enterprise-grade inference
llama.cppNoneLocal, offline, privacy-first
LM StudioNoneGUI-based local model management
SAMAPI KeyIntegration with SAM ecosystem

Variant providers target the same backends but with different billing endpoints or quotas:

VariantDifference
MiniMax Token PlanMiniMax Token Plan subscription (quota-based, separate billing)
Z.AI Coding PlanZ.AI /api/coding/paas/v4 endpoint with coding-plan subscription pricing

See docs/PROVIDERS.md for setup instructions for each provider.

Screenshots

CLIO's Simple User Interface (1/2)

CLIO's Simple User Interface (2/2)

Quick Start

Check Dependencies

./check-deps  # Verify all required tools are installed

CLIO requires standard Unix tools (git, curl, perl 5.32+, etc.). See docs/DEPENDENCIES.md for details.

Install

Homebrew (macOS)

brew tap SyntheticAutonomicMind/homebrew-SAM
brew install clio

Docker (Any Platform)

docker run -it --rm \
    -v "$(pwd)":/workspace \
    -v clio-auth:/root/.clio \
    -w /workspace \
    ghcr.io/syntheticautonomicmind/clio:latest \
    --new

Manual

git clone https://github.com/SyntheticAutonomicMind/CLIO.git
cd CLIO && sudo ./install.sh

For detailed options, see docs/INSTALLATION.md.

Configure AI Provider

GitHub Copilot (Recommended - no config needed)

./clio
: /api login
# Follow browser prompts to authorize

Other Providers

./clio
: /api set provider openai
: /api set key YOUR_API_KEY
: /config save

First Prompt

Start with something grounded in a real repository:

Read this project and tell me how configuration is loaded.

Or give CLIO a real bug to investigate:

Find the bug causing the login endpoint to return 500 when the session is expired.

Start Using CLIO

./clio --new          # New conversation
./clio --resume       # Resume last session
./clio --debug        # Debug mode
./clio --enable file_operations  # Restrict to specific tools
./clio --disable web_operations  # Block specific tools

Requirements

  • macOS 10.14+ or Linux (any modern distribution)
  • Perl 5.32+ (included with most systems)
  • Git (for version control operations)
  • ANSI-compatible terminal

Privacy and Security

CLIO runs with defense-in-depth security:

  • Secret redaction - API keys, tokens, and credentials are stripped from AI context before it reaches the model
  • Command analysis - Shell commands are classified by risk level (network, credential access, destructive) and require your approval for high-risk operations
  • Path authorization - File access outside the project directory requires your permission
  • Sandbox mode - --sandbox blocks web/remote/agent access and restricts file operations to the project directory
  • Container isolation - clio-container provides full OS-level isolation via Docker
  • Invisible character filtering - Unicode-based prompt injection attacks are blocked automatically

See docs/SECURITY.md and docs/SANDBOX.md for details.

Part of the Ecosystem

CLIO is part of Synthetic Autonomic Mind - a family of open source AI tools:

  • SAM - Native macOS AI assistant with voice control, document analysis, and image generation
  • ALICE - Local Stable Diffusion server with web interface and OpenAI-compatible API
  • SAM-Web - Access SAM from iPad, iPhone, or any browser

Documentation

DocumentWhat You'll Find
User GuideComplete usage guide with examples
FeaturesEvery feature explained in detail
InstallationGetting started with CLIO
ProvidersAI provider configuration guide
DependenciesSystem requirements and verification
Sandbox ModeSecurity isolation options
ArchitectureSystem design and internals
MemoryHow CLIO remembers and learns across sessions
Developer GuideContributing and extending CLIO
Remote ExecutionDistributed AI workflows
Multi-AgentParallel agent coordination
MCP IntegrationModel Context Protocol support
Custom InstructionsPer-project AI customization
SecuritySecurity model and secret redaction
PerformanceBenchmarks and optimization
Style GuideColor themes and customization
AutomationCLIO-helper daemon and CI/CD integration
Puppeteer ModeMulti-project orchestration

License

GPL-3.0-or-later - See LICENSE for details. · Created by Andrew Wyatt (Fewtarius) · syntheticautonomicmind.org · GitHub

Support