Rai — Run AI instructions in your shell, scripts, and CI/CD pipelines.

Quick demo ↓

In your terminal

Run AI in your shell — pipe and chain it like any Unix tool, from quick one-liners to structured markdown runbooks.

                $ rai "whois github.com"
                Registrar: MarkMonitor Inc.
              
                $ ls -al | rai "count all file size"
                Total: 42 files, 1.2 MB combined
              
                $ rai --model kimi-k2.5 "FIX_TODO.md" ~/my-project
                Fixed 5 TODOs across 3 files. See diff below.

In your scripts

Supercharge your project workflows with AI-powered scripts.

              {
                "scripts": {
                  "test": "npm run test",
                  "check": "rai rules-check.md & rai security.md"
                }
              }
            

In CI/CD

Drop any AI steps into your build pipeline.

              steps:
                - run: npm install
                - run: npm run build
                - run: rai translate.md
                - run: npm run deploy
            

Why rai ↓

Open source & lightweight

Single binary, no runtime dependencies. Installs in seconds and starts instantly.

Unix-native

Supports arguments, piping, and chaining. Fits seamlessly into your existing shell scripts and workflows.

No lock-in

Choose your own provider and model — from OpenAI, Anthropic, and Google to local LLMs via Ollama.

Privilege control

Fine-grained tool permissions and rule-based guardrails keep AI actions safe and scoped.

Good for

Quick AI tasks, script automation, and CI/CD steps — use it like grep or curl.

Not designed for

Complex multi-hour tasks or long-running persistent AI agents.

Usage ↓

Install

            # via cargo
            $ cargo install rai-cli
             
            # or via curl
            $ curl -sSL https://appmakes.github.io/Rai/install.sh | sh
            # uninstall (curl install only)
            $ curl -sSL https://appmakes.github.io/Rai/uninstall.sh | sh
          

Setup & configure

First-time setup picks your provider, API key, and default model. Revisit anytime with rai config.

            $ rai start
            ? Select provider: OpenAI
            ? API key: sk-••••••••
            ? Default model: gpt-4o
            ✓ Config saved to ~/.config/rai/config.toml
          

Common flags

-m, --model Override model per run

-y, --yes Auto-approve tool calls

-s, --silent No follow-up input

--no-tools Disable tool calling

--bill Print token & cost summary

--profile <name> Use a named config profile

--<variable> <value> Pass custom variables to task files

Task files

Run prompts directly, or write them as markdown files for complex, reusable tasks. Task files support frontmatter for model config and {{ variables }} for templating.

            # Run a prompt directly
            $ rai "test my network speed"
             
            # Or run a task file
            $ rai review.md
          

Example task file with variables:

            # review.md
            ---
            model: gpt-4o
            args:
              - target
              - language
              - focus?
            ---
             
            Review {{ target }} written in {{ language }}.
            Focus on {{ focus }} if specified.
          

Pass variables as positional or named args:

            $ rai review.md src/main.rs rust
            $ rai review.md --target src/ --language rust --focus security
          