GitHub - tgalal/promptcmd: GenAI prompts as runnable programs

Quick Start | Documentation | Examples

promptcmd is a manager and executor for programmable prompts. Define a prompt template once, then execute it like any other terminal command, complete with argument parsing, --help text, and stdin/stdout integration.

Unlike tools that require you to manually manage prompt files or rely on implicit tool-calling, promptcmd gives you explicit control over what data your models access. Build compositional workflows by nesting prompts, executing shell commands within templates, and piping data through multi-step AI pipelines.

Key Features

Prompts as CLI Commands

Create a .prompt file, enable it with promptctl, and execute it like any native tool.

$ promptctl create bashme_a_script_that
$ bashme_a_script_that renames all files in current directly to ".backup"

Execute in Remote Shells

Prepend SSH commands with promptctl, your prompts magically appear in your remote shell sessions.

$ promptctl ssh user@server
server $ bashme_a_script_that renames all files in current directly to ".backup"

Local and Remote Provider Support

Use your Ollama endpoint or configure an API key for OpenAI, OpenRouter, Anthropic, Google, or MiniMax. Swap between them with ease.

$ promptctl create render-md
$ cat README.md | render-md -m openai
$ cat README.md | render-md -m ollama/gpt-oss:20b

Group and Load Balancing

Distribute requests across several providers with equal or weighted distribution for cost optimization.

# config.toml
[groups.balanced]
providers = ["openai", "google"]

$ cat README.md | render-md -m balanced

Caching

Cache responses for a configured amount of time for adding determinism in pipelines and more efficient token consumption.

# config.toml
[providers.openai]
cache_ttl = 60 # number of seconds

Set/Override during execution:

$ cat README.md | render-md --config-cache-ttl 120

Custom Models with Character

Use Variants to define custom models with own personality or specialization in tasks:

[providers.anthropic]
api_key = "sk-xxxxx"
model = "claude-sonnet-4-5"

[providers.anthropic.glados]
system = "Use sarcasm and offending jokes like the GlaDoS character from Portal."

[providers.anthropic.wheatley]
system = "Reply as if you are Wheatley from Portal."

$ tipoftheday -m glados
$ tipoftheday -m wheatley

Quick Start

Install

Linux/macOS:

curl -LsSf https://installer.promptcmd.sh | sh

macOS (Homebrew):

brew install tgalal/tap/promptcmd

Windows (PowerShell):

powershell -ExecutionPolicy Bypass -c "irm https://installer-ps.promptcmd.sh | iex"

Configure API Keys

Configure your API keys by editing config.toml:

Find your provider's name, e.g., for anthropic:

[providers.anthropic]
api_key = "sk-ant-api03-..."

Alternatively, you can set the keys via Environment Variables:

PROMPTCMD_ANTHROPIC_API_KEY="your_api_key"
PROMPTCMD_OPENAI_API_KEY="your_api_key"
PROMPTCMD_ANTHROPIC_API_KEY="your_api_key"
PROMPTCMD_OPENROUTER_API_KEY="your_api_key"
PROMPTCMD_MINIMAX_API_KEY="your_api_key"

Create Your First Prompt

Create a summarize.prompt file:

promptctl create summarize

Insert the following:

---
model: anthropic/claude-sonnet-4-5
input:
  schema:
    words?: integer, Summary length in words
---
Summarize the following text{{#if words}} in {{words}} words{{/if}}:

{{STDIN}}

Enable and use it:

# Enable as a command
promptctl enable summarize

# Use it
cat article.txt | summarize
echo "Long text here..." | summarize --words 10

# Auto-generated help
summarize --help

That's it. Your prompt is now a native command.

Documentation

Full documentation available at: docs.promptcmd.sh

Examples

Browse the Examples directory or visit https//promptcmd.sh/lib for interactive viewing.

License

GPLv3 License - see LICENSE file for details