AIPACK - Run, Build, and Share AI Packs

Checkout 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.

• jc@coder Pack Demo
• AIPACK Introduction
• AIPACK Playlist

Quick Start

Install

For now, the install requires building it directly from source via Rust. Works great on all 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 (about Mid / End of March)

Run

# In terminal go to your projct
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 proof read 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 packname of this name
aip run @proof -f ./README.md

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

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

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

jc@coder

• You can install jc@coder with aip install jc@coder, and then
• Run it with aip run jc@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 usage (at least while you are learning about aipack).

IMPORTANT 2: Make sure to have your API_KEY in an environment variable (on Mac, there is an 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 all the 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 tuneup. Pack install test and other refactoring
  • 2025-03-02 (v0.6.4) - Fixes, and now support first repo pack aip install jc@coder
  • 2025-02-28 (v0.6.3) - aip pack .., aip instal local..., ai_response.price_usd, and more
  • 2025-02-26 (v0.6.0) - BIG UPDATE - to AIPACK, now with pack support (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 cool 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 from Mac/Unix-like systems, which might not function correctly in the Windows bat command line.
  • Full Windows local path support is in development.
  • RECOMMENDATION: Use PowerShell or WSL on Windows. Please log issues if small changes can accommodate Windows PowerShell/WSL.

• Thanks to

  • Stephane Philipakis, a top 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
  • A .aip Agent file is just a Markdown File with sections for each stage of the agent 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 name proof
  • namespace demo
  • agent file main.aip
  • Full path ~/.aipack-base/pack/installed/demo/proof/main.aip
  • You can pass input to your agent with
    • -f "path/with/optional/**/glob.*" -f "README.md (then the lua code will get a {path = .., name =..} FileMeta type of structure as input)
    • -i "some string" -i "another input" (then the lua code will get those strings as input)
    • Each input will be one run of the agent.
• aip run some/path/to/agent
  • can end with .aip in this case direct file run
  • if no .aip extension, then,
    • ...agent.aip will be executed if exists
    • or ...agent/main.aip will be executed if exists
• aipack agents are simple .aip files that can be placed anywhere on disk.
  • e.g., aipack run ./my-path/to/my-agent.aipack ...
• Multi AI Provider / Models - aipack uses genai and therefore supports OpenAI, Anthropic, Gemini, Groq, Ollama, Cohere, and more to come.
• 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 command global data to scope (the "map" of the map/reduce capability).
# Data	Lua	Gather additional data per input and return it for the next stages.
# System	Handlebars	Customize the prompt with the data and before_all data.
# Instruction	Handlebars	Customize the prompt with the data and before_all data.
# Assistant	Handlebars	Optional for special customizations, such as the "Jedi Mind Trick."
# Output	Lua	Processes the ai_response from the LLM. Otherwise, ai_response.content will be output to the terminal.
# After All	Lua	Called with inputs and outputs for post-processing after all inputs are completed.

• # Before All / # After All can be considered as the map/reduce of the agent, and these will be run before and after the input processing.

more info on stages

aipack doc

See the aipack documentation at _init/doc/README.md (With Lua modules doc)

You can also run the ask-aipack agent.

# IMPORTANT - Make sure you have the `OPENAI_API_KEY` or the key of your model in your environment
aip run core@ask-aipack
# prompt file will be at `.aipack/.prompt/core@ask-aipack/ask-prompt.md`