Andi AIRun

Run AI prompts like programs. Executable markdown with shebang, Unix pipes, and output redirection. Supports multiple runtimes (Claude Code, Codex CLI) with cross-cloud provider switching and any-model support — free local or 100+ cloud models.

# Claude Code: any model or provider
ai                                        # Regular Claude subscription (Pro, Max)
ai --aws --opus --team --resume           # Resume chats on AWS w/ Opus 4.6 + Agent Teams
ai --ollama --bypass --model qwen3-coder  # Ollama local model with bypassPermissions set

# Codex CLI: OpenAI's coding agent
ai --codex                                # Codex with gpt-5.4 (default)
ai --codex --high                         # Codex with gpt-5.4 (flagship)
ai --codex --ollama                       # Codex with local Ollama models

# Run prompts like programs (works with any runtime)
ai --azure --haiku script.md
ai --codex script.md

# Script automation
cat data.json | ./analyze.md > results.txt

Choose your runtime — Claude Code or Codex CLI — and switch between clouds + models: AWS Bedrock, Google Vertex, Azure, Vercel, Anthropic API, OpenAI API. Supports free local models (Ollama, LM Studio) and 100+ alternate cloud models via Vercel AI Gateway or Ollama Cloud. Swap and resume conversations mid-task to avoid rate limits and keep working.

What it does:

Multiple runtimes: Claude Code and Codex CLI with a single ai command (--cc, --codex)
Executable markdown with #!/usr/bin/env ai shebang for script automation
Unix pipe support: pipe data into scripts, redirect output, chain in pipelines
Cross-cloud provider switching: use Claude on AWS, Vertex, Azure, Anthropic API, or Codex on OpenAI, Azure OpenAI, OpenRouter + switch mid-conversation to bypass rate limits
Model tiers: --opus/--high, --sonnet/--mid, --haiku/--low — maps to each runtime's models
Cross-interpreter effort control: --effort low|medium|high|max
Session continuity: --resume picks up your previous chats with any model/provider
Non-destructive: plain claude and codex always work untouched as before

From Andi AI Search. Star this repo if it helps!

Latest: Codex CLI support (--codex), cross-interpreter effort levels (--effort), tool profiles (--profile). Script variables, live streaming, Agent Teams, Opus 4.6, local models (Ollama, LM Studio), persistent defaults, 100+ cloud models via Vercel. See CHANGELOG.md.

Quick Start

Supported Platforms:

macOS 13.0+
Linux (Ubuntu 20.04+, Debian 10+)
Windows 10+ via WSL

Prerequisites: At least one runtime installed — Claude Code or Codex CLI

# Install a runtime (one or both)
curl -fsSL https://claude.ai/install.sh | bash   # Claude Code (Anthropic)
npm install -g @openai/codex                      # Codex CLI (OpenAI)

# Install Andi AIRun
git clone https://github.com/andisearch/airun.git
cd airun && ./setup.sh

You can now run any markdown file as an AI script:

# Create an executable prompt
cat > task.md << 'EOF'
#!/usr/bin/env ai
Analyze my codebase and summarize the architecture.
EOF

chmod +x task.md
./task.md                         # Runs with your Claude subscription

Or run any markdown file directly:

ai task.md

Pipe data and redirect output (Unix-style automation):

cat data.json | ./analyze.md > results.txt    # Pipe in, redirect out
git log -10 | ./summarize.md                  # Feed git history to AI
./generate.md | ./review.md > final.txt       # Chain scripts together

Run scripts from the web (installmd.org support):

curl -fsSL https://andisearch.github.io/ai-scripts/analyze.md | ai
echo "Explain what a Makefile does" | ai         # Simple prompt

Minimal alternative: If you just want basic executable markdown without installing this repo, add a ai script to your PATH:

#!/bin/bash
claude -p "$(tail -n +2 "$1")"

This works for simple prompts but lacks provider switching, model selection, stdin piping, output formats, and session isolation. (credit: apf6)

Commands

| Command | Description | |---------|-------------| | ai / airun | Universal entry point - run scripts, switch providers | | ai update | Update AI Runner to the latest version | | ai-sessions | View active AI coding sessions | | ai-status | Show current configuration and provider status |

Running ai with no flags matches your claude defaults — if you're logged in with a subscription, ai uses it. Your environment is automatically restored on exit. Add provider flags to switch, or use ai --aws --opus --set-default to save your preferred provider and model for future runs.

Usage Examples

# Run a markdown script (auto-detects runtime + provider)
ai task.md

# Choose your runtime
ai --cc                           # Claude Code (default if installed)
ai --codex                        # Codex CLI (OpenAI)

# Claude Code providers
ai --aws                          # AWS Bedrock
ai --vertex                       # Google Vertex AI
ai --apikey                       # Anthropic API
ai --azure                        # Microsoft Azure Foundry
ai --vercel                       # Vercel AI Gateway
ai --pro                          # Claude Pro/Max subscription

# Codex CLI providers
ai --codex                        # OpenAI API (default)
ai --codex --azure                # Azure OpenAI (via config.toml)
ai --codex --profile openrouter   # OpenRouter (via config.toml profile)

# Local models (work with both runtimes)
ai --ollama                       # Ollama with Claude Code
ai --codex --ollama               # Ollama with Codex CLI
ai --lmstudio                     # LM Studio (MLX, Apple Silicon)

# Model tiers (map to each runtime's best models)
ai --opus task.md                 # Claude: Opus 4.6 / Codex: gpt-5.4
ai --sonnet task.md               # Claude: Sonnet 4.6 / Codex: gpt-5.3-codex (mid tier)
ai --haiku task.md                # Claude: Haiku 4.5 / Codex: gpt-5.4-mini
ai --codex --high task.md         # Codex with gpt-5.4

# Effort level (cross-interpreter reasoning control)
ai --effort high task.md          # Claude Code: deeper reasoning
ai --codex --effort max task.md   # Codex: maximum reasoning (xhigh)

# Stream output in real-time
ai --live --skip task.md

# Suppress --live status for CI/CD (clean stdout only)
ai --quiet ./live-script.md > output.md

# Live output + file redirect (narration to console, clean content to file)
./live-report.md > report.md

# Override script variables (--topic, --style match declared vars: names)
./summarize-topic.md --live --topic "the fall of rome" --style "peter griffin"

# Resume last conversation
ai --aws --resume

# Save runtime + provider + model as default
ai --codex --high --set-default   # Always use Codex + gpt-5.4
ai --aws --opus --set-default     # Always use Claude Code + AWS + Opus
ai --clear-default                # Remove saved default

# Smart auto permissions (AI classifier for Claude Code, sandbox for Codex)
ai --auto task.md

# Enable agent teams (Claude Code, experimental, interactive only)
ai --team                         # Auto display mode
ai --aws --opus --team            # Teams with AWS Bedrock + Opus

Features

Executable Markdown

Create markdown files with prompts that run directly via shebang:

#!/usr/bin/env ai
Summarize the architecture of this codebase.

#!/usr/bin/env -S ai --aws
Use AWS Bedrock to analyze this code.

#!/usr/bin/env -S ai --codex --high
Use Codex CLI with the flagship model to review this code.

#!/usr/bin/env -S ai --opus --live
Review this PR for security issues. Stream output in real-time.

Scripts that write files or run commands need a permission mode:

#!/usr/bin/env -S ai --skip
Run ./test/automation/run_tests.sh and report results.

(--skip is a shortcut for --dangerously-skip-permissions. See also --bypass for --permission-mode bypassPermissions.)

#!/usr/bin/env -S ai --auto
Run tests and fix any issues found.

(--auto uses an AI classifier (Claude Code) or sandbox (Codex) to auto-approve safe actions.)

#!/usr/bin/env -S ai --allowedTools 'Bash(npm test)' 'Read'
Run the test suite and report results. Do not modify any files.

(--allowedTools is a Claude Code flag, passed through by AI Runner.)

Usage:

chmod +x task.md
./task.md                          # Execute directly (uses shebang flags)
ai --vercel task.md                # Override: use Vercel instead
ai --opus task.md                  # Override: use Opus instead

Tip: Use #!/usr/bin/env -S (with -S) to pass flags in the shebang line. Standard env only accepts one argument, so #!/usr/bin/env ai --aws won't work — you need -S to split the string.

Flag precedence: CLI flags > shebang flags > saved defaults. Running ai --vercel task.md overrides the script's shebang provider. Shebang fla

Airun

Install / Use

README