AIPACK - Run, Build, and Share AI Packs

Check out the site: https://aipack.ai for more information and links.

Open-source Agentic Runtime to run, build, and share AI Packs.

• Supports all major AI providers and models.
• Efficient and small (< 20MB), with zero dependencies.
• Built in Rust using Lua for embedded scripting (small and efficient).
• Runs locally, completely IDE-agnostic.
• Or in the cloud—server or serverless.

• Video: jc@coder (now pro@coder) Pack Demo
• Video: AIPACK Introduction
• Video: AIPACK Playlist
• AIPACK Substack at news.aipack.ai

Quick Start

Install

For now, installation requires building directly from source via Rust. Works on all major OSes.

• Install Rust: https://www.rust-lang.org/tools/install
• For now, install with cargo install aipack

NOTE: Ironically, while the binary is relatively small (<20MB with batteries included), the build process can take up quite a bit of space. However, Cargo should clean it up afterward. Binaries and installation instructions will be available at https://aipack.ai

DISCLAIMER: For now, v0.6.x, AIPACK works on Linux & Mac, and requires WSL on Windows.

IMPORTANT: Proper Windows support is coming sometime in v0.6.x and definitely by v0.7.x (around Mid/End of March).

Run

# In the terminal, go to your project
cd /path/to/my/project/

# Initialize workspace .aipack/ and ~/.aipack-base
aip init

# Make sure to export the desired API key
export OPENAI_API_KEY    = "sk...."
export ANTHROPIC_API_KEY = "...."
export GEMINI_API_KEY    = "..."
# For more keys, see below

# To proofread your README.md (namespace: demo, pack_name: proof)
aip run demo@proof -f ./README.md

# You can just use @pack_name if there is no other pack with this name
aip run @proof -f ./README.md

# To do some code crafting (will create `_craft-code.md`)
aip run demo@craft/code

# Or run your .aip file (you can omit the .aip extension)
aip run path/to/file.aip

# This is a good agent to run to ask questions about aipack
# It can even generate aipack code
aip run core@ask-aipack
# The prompt file will be at `.aipack/.prompt/core@ask-aipack/ask-prompt.md`

pro@coder

• You can install pro@coder with aip install pro@coder, and then
• Run it with aip run pro@coder or aip run @coder if you don't have any other @coder pack in a different namespace.

This is the agent I use every day for my production coding.

IMPORTANT 1: Make sure everything is committed before use (at least while you are learning about aipack).

IMPORTANT 2: Make sure to have your API_KEY set as an environment variable (on Mac, there is experimental keychain support).

OPENAI_API_KEY
ANTHROPIC_API_KEY
GEMINI_API_KEY
XAI_API_KEY
DEEPSEEK_API_KEY
GROQ_API_KEY
COHERE_API_KEY

Info

• Website: https://aipack.ai

• AIPACK Overview Video

• Preview 'devai' intro video for v0.5

• Built on top of the Rust genai library, which supports many top AI providers and models (OpenAI, Anthropic, Gemini, DeepSeek, Groq, Ollama, xAI, and Cohere).

• Top new features: (see full CHANGELOG)

  • 2025-03-02 (v0.6.7) - Fixes and tune-up. Pack install test and other refactoring
  • 2025-03-02 (v0.6.4) - Fixes, and now supports the first repo pack aip install jc@coder
  • 2025-02-28 (v0.6.3) - aip pack .., aip install local..., ai_response.price_usd, and more
  • 2025-02-26 (v0.6.0) - BIG UPDATE - to AIPACK, now with pack support (e.g., aip run demo@craft/code)
  • 2025-02-22 (v0.5.11) - Huge update with parametric agents and coder (more info soon)
  • 2025-01-27 (v0.5.9) - DeepSeek distill models support for Groq and Ollama (local)
  • 2025-01-23 (v0.5.7) - aipack run craft/text or aipack run craft/code (example of new agent module support)
  • 2025-01-06 (v0.5.4) - DeepSeek deepseek-chat support
  • 2024-12-08 (v0.5.1) - Added support for xAI

• WINDOWS DISCLAIMER:

  • This CLI uses a path scheme similar to Mac/Unix-like systems, which might not function correctly in the Windows cmd.exe (Command Prompt) or traditional batch files (.bat).
  • Full Windows local path support is in development.
  • RECOMMENDATION: Use PowerShell or WSL on Windows. Please log issues if minor changes can improve support for Windows PowerShell/WSL.

• Thanks to

  • Stephane Philipakis, a key aipack collaborator.
  • David Horner for adding Windows support for Open Agent (with VSCode) (#30)
  • Diaa Kasem for --non-interactive/--ni mode (#28)

How it works

• One Agent == One Markdown
  • An .aip agent file is just a Markdown file with sections for each stage of the agent's processing.
  • See below for all the possible stages.
• aip run demo@proof -f "./*.md"
  • will run the installed agent file main.aip in the
  • pack named proof
  • namespace demo
  • agent file main.aip
  • Full path ~/.aipack-base/pack/installed/demo/proof/main.aip
  • You can pass input to your agent using:
    • -f "path/with/optional/**/glob.*" -f "README.md" (the Lua code will receive a {path = .., name =..} FileMeta-like structure as input)
    • -i "some string" -i "another input" (the Lua code will receive these strings as input)
    • Each input triggers one run of the agent.
• aip run some/path/to/agent
  • If the path ends with .aip, it's treated as a direct file run.
  • If there is no .aip extension, then:
    • .../agent.aip will be executed if it exists.
    • or .../agent/main.aip will be executed if it exists.
• aipack agents are simple .aip files that can be placed anywhere on disk.
  • e.g., aipack run ./my-path/to/my-agent.aip ...
• Multi AI Provider / Models - aipack uses genai and therefore supports providers like OpenAI, Anthropic, Gemini, Groq, Ollama, Cohere, and more.
• Lua is used for all scripting (thanks to the great mlua crate).
• Handlebars is used for all prompt templating (thanks to the great Rust native handlebars crate).

Multi Stage

A single aipack file may comprise any of the following stages.

Stage	Language	Description
# Before All	Lua	Reshape/generate inputs and add global data to the command scope (the "map" part of map/reduce).
# Data	Lua	Gather additional data per input and return it for the next stages.
# System	Handlebars	Customize the system prompt using data from # Before All and # Data.
# Instruction	Handlebars	Customize the instruction prompt using data from # Before All and # Data.
# Assistant	Handlebars	Optional for special customizations, such as the "Jedi Mind Trick."
# Output	Lua	Processes the ai_response from the LLM. If not defined, ai_response.content is output to the terminal.
# After All	Lua	Called with inputs and outputs for post-processing after all input processing is complete (the "reduce" part of map/reduce).

• # Before All and # After All act like the map and reduce steps, running before and after the main input processing loop, respectively.

more info on stages

aipack doc

See the aipack documentation at core/doc/README.md (with the Lua modules doc).

You can also run the ask-aipack agent.

# IMPORTANT: Make sure you have the `OPENAI_API_KEY` (or the key for your desired model) set in your environment
aip run core@ask-aipack
# The prompt file will be at `.aipack/.prompt/core@ask-aipack/ask-prompt.md`