Hal9: Create and Share Generative Apps

Create and deploy generative (LLMs and diffusers) applications (chatbots and APIs) in seconds.

• Open: Use any model (OpenAI, Llama, Groq, MidJourney) and any library like (LangChain, DSPy).
• Intuitive: No need to learn app frameworks (Flask), simply use input() and print(), or write file to disk.
• Scalable: Engineers can integrate your app with scalable technologies (Docker, Kubernetes, etc)
• Powerful: Using an OS process (stdin, stdout, files) as our app contract, enables long-running agents, multiple programming languages, and complex system dependencies.

Focus on AI (RAG, fine-tuning, alignment, training) and skip engineering tasks (frontend development, backend integration, deployment, operations).

Getting started

Create and share a chatbot in seconds as follows:

pip install hal9

hal9 create chatbot
hal9 deploy chatbot

Notice that deploy needs a HAL9_TOKEN environment variable with an API token you can get from hal9.com/devs. You can use this token to deploy from your local computer, a notebook or automate from GitHub.

HAL9_TOKEN=H9YOURTOKEN hal9 deploy chatbot --name my_first_chatbot

As easy as that you have created your first chatbot!

The code inside /chatbot/app.py contains a "Hello World" chatbot that reads the user prompt and echos the result back:

prompt = input()
print(f"Echo: {prompt}")

We designed this package with simplicity in mind, the job of the code is to read input and write output, that's about it. That said, you can create chatbots that use LLMs, generate images, or even use tools that connect to databases, or even build websites and games!

Creation

By default hal9 create defaults to the --template echo template, but you can choose different ones as follows:

hal9 create chatbot-openai --template openai
hal9 create chatbot-groq --template groq

A template provides ready to use code with specific technologies and use cases. Is very popular to use OpenAI's ChatGPT-like template with --template openai, the code generated will look as follows:

import hal9 as h9
from openai import OpenAI

messages = h9.load("messages", [])
prompt = h9.input(messages = messages)

completions = OpenAI().chat.completions.create(model = "gpt-4", messages = messages, stream = True)

h9.complete(completions, messages = messages)
h9.save("messages", messages, hidden = True)

The Learn section explain in detail how this code works, but will provide a quick overview. The hal9 package contains a helper functions to simplify your generative AI code. You can choose to not use hal9 at all and use input() and print() statements yourself, or even sue tools like langchain. The h9.load() and h9.save() functions load and save data across chat sessions, our platform is stateless by default. The h9.input() function is a slim wrapper over input() that also stores the user input in the messages. Then h9.complete() is a helper function to help parse the completion results and save the result in messages. That's about it!

Development

To make changes to your project, open chatbot/ in your IDE and modify chatbot/app.py.

You can then run your project as follows:

hal9 run chatbot

If you customized your template with --template make sure to set the correct key, for example, if you are using the OpenAI template use for Linux or macOS:

export OPENAI_KEY=YOUR_OPENAI_KEY.

For Windows use:

set OPENAI_KEY=YOUR_OPENAI_KEY.

For more information on obtaining and using your OpenAI API key, please refer to the OpenAI API Key documentation.

You can then run your application locally with:

hal9 run chatbot

This command is just a convenience wrapper to running the code yourself with something like python app.py.

Deployment

The deploy command will prepare for deployment your generative app.

For example, you can prepare deployment as a generative app (Hal9). We have plans to also provide deployment to Docker and the open source community can expand this even further.

hal9 deploy chatbot --target hal9

Each command is tasked with preparing the deployment of your project folder. For example, --target docker should create a Dockerfile file that gets this project ready to run in cloud containers.

For personal use, --target hal9 supports a free tier at hal9.com; enterprise support is also available to deploy with --target hal9 --url hal9.yourcompany.com