instrukt (0.5.0)
Installation
pip install --index-url About this package
Build and instruct AI agents to perform tasks and query infromation in a keyboard driven terminal UI.
Instrukt
An AI Commander at your fingertips
NOTE: This is an alpha release, expect bugs and API changes.
Introduction
Instrukt is a terminal-based environment for AI productivity and development. It offers a platform where users can:
- 🤖 Create and instruct modular AI agents
- 🗃️ Generate document indexes for question-answering
- 🧰 Create and attach tools to any agent
Instrukt agents are simple drop-in Python packages that can be extended, shared with others, attached to tools, and augmented with document indexes. Instruct them in natural language and, for safety, attach them to inside secure containers (currently implemented with Docker) to perform tasks in their dedicated, sandboxed space 🛡️.
Instrukt is designed with with an easy interface and API for composing and using agents. It is built with LangChain, extending its base agents and tools.
For agent developers, it offers a builtin IPython console for on-the-go debugging and introspection and allows for quick development and testing of Langchain agents in an interactive environment.
Built with: Langchain, Textual, Chroma
Consulting Services: Need help with Langchain or AI development/deployment ? You can reach out to me at contact@blob42.xyz
TOC
- Usage
- Features
- Supported Platforms
- LLM Models
- Document Indexes and Question-Answering
- Roadmap
- Contributing
- Patreon
- Social
- Vision
- License
Usage
Setup
note: the package is not yet released on PyPi
From pre-built package:
- download and install the released package with
pip install package_name[tools].
From source:
- Make sure the latest version of
poetryis installed. - Set your virtualenv
- Clone the repository
- Run
poetry install -E tools --with dev - This will install Instrukt including extra tools for agents.
See the installation guide for more details
Requirements:
- Export
OPENAI_API_KEYenvironment variable with your OpenAI API key. - Use a modern terminal that supports 24-bit color and mouse input.
Running:
- Type
instrukt. - On the first run, it will create a config file in
~/.config/instrukt/instrukt.yml
From source
-
If you did not install the dependencies using poetry make sure to install textual dev dependencies with:
pip install textual[dev] -
From the project's root run:
textual run instrukt.app:InstruktApp
Default Agents:
Chat Q&A: A simple conversational agent. Attach document indexes and tools to create a question-answering agent. It also serves as an example for making custom agents.
Demo: The same agent used in the demo video.
Features
💻 Keyboard and Mouse Terminal Interface:
- A terminal-based interface for power keyboard users to instruct AI agents without ever leaving the keyboard.
- Mouse support and Rich colorful agent outputs such as markdown and source
code thanks to the Textual TUI library. - Run Instrukt in bare metal and access it remotely via SSH and/or terminal
multiplexer.
🤖 Custom AI Agents:
- Design your own agents and custom tools.
- Agents are simple python packages can be shared and loaded by other users.
🔧 Tools:
- Use the pre-defined toolset or design your own tools.
- Connect or disconnect tools to agents on-the-go, tailoring your AI workflows to your needs.
📚 Vectorstore Indexes:
Create indexes over your data, attach them to agents, and use them for question-answering.
⚡ Prompt Console :
- Integrated REPL-Prompt for quick interaction with agents, and a fast feedback loop for development and testing.
- Add your own custom commands to automate repetitive tasks.
- Builtin prompt/chat history to quickly edit/correct messages.
🐦 LangChain:
- Leverage the LangChain ecosystem to automate anything.
- Extensible API for integrating with other frameworks.
🔬 Developer Console:
Debug and introspect agents using an in-built IPython console.
🛡️ Secure Containers:
- Run agents inside secure docker containers for safety and privacy.
- Use gVisor runtime for a full isolation of the agent.
note: The docker agent is only available to Patreon supporters as an early preview.
Document Indexes and Question-Answering
- Indexes are created using OpenAI embeddings.
- They are stored locally using the Chroma embedding database.
- You create and manage indexes using the Index Management UI press capital
I - An index can be attached to any agent as a retrieval tool using the
indexmenu
from the agent's window. - Once an index is attached you can do question-answering with your agent and it will
automatically lookup your index whenever its name is mentioned.
note: Indexing is currently using naive document chunking/splitting heuristics without any optimizations. The quality of document retrieval and agent's answers can be greatly improved in future iterations.
Supported platforms:
- Linux/Mac.
- Windows tested under WSL2.
LLM Models
- Currently only OpenAI supported.
- Using private local models is the next priority.
Roadmap
- Private local models
- Use text-generation-webui API
- Emulate PrivateGPT
- Use a self hosted go-skynet/LocalAI
- Local embeddings
- wip: HF SetenceTransformers or other embeddings models.
- Instructor Embeddings
- Documentation
- Creating agents
- Creating tools
- Indexing and chat with documents and source code.
- Example use cases
- Tutorials for users non familiar with AI/LLM terms.
Contributing
Any contribution, feedback and PR is welcome !
You can help with:
- Testing and creating Issues for bugs or features that would be useful.
- If you have technical skills, you are welcome to create a PR.
- If you don't have technical skills you can help with documentation, adding examples
and tutorials or create new user stories.
Patreon
By becoming a patron, you will help me continue committing time to the development of Instrukt and bring to life all the planned features. Check out the Patreon page for more details on the rewards for early supporters.
Social
Join the Discord server to keep updated on the progress or ask for help.
Vision
I believe that AI should be accessible to everyone and not a walled garden for big corporations and SaaS services.
I hope my contributions will help create tools that empower users without compromising their freedoms. The long-term goal for Instrukt is to make it usable with minimal reliance on external services, giving users the choice to opt for local models and self-hosted services.
License
Copyright (c) 2023 Chakib Ben Ziane. All Rights Reserved.
Instrukt is licensed with a AGPL license, in short this means that it can be used by anyone for any purpose. However, if you decide to make a publicly available instance your users are entitled to a copy of the source code including all modifications that you have made (which needs to be available trough an interface such as a button on your website), you may also not distribute this project in a form that does not contain the source code (Such as compiling / encrypting the code and distributing this version without also distributing the source code that includes the changes that you made. You are allowed to distribute this in a closed form if you also provide a separate archive with the source code.).