/beebot

An Autonomous AI Agent that works

Primary LanguagePythonMIT LicenseMIT

BeeBot

BeeBot is your personal worker bee, an Autonomous AI Assistant designed to perform a wide range of practical tasks autonomously.

BeeBot Mascot

Status

Development of BeeBot is currently on hold. I've decided that LLMs as they are now (late 2023) aren't up to the task of generalized autonomous AI. I will revive the project if either:

  • LLMs get significantly better at structured thinking, reliable outcomes, and obeying instructions
  • I can develop or fine tune a custom model which is trained specifically for Autonomous AI
  • I figure out a particular subset of tasks that BeeBot is acceptably good at that I can focus on. (Hint: It's not coding)

Check back here, hopefully this will get re-started.

Features

  • Tool selection via AutoPack and the ability to acquire more tools during task execution
  • Built-in persistence
  • REST API conforming to the e2b standard.
  • A websocket server to publish all events that occur within BeeBot
  • Swappable filesystem emulation so that files can be stored in-memory, on-disk, or in a database
  • A Web UI for managing your tasks (coming very soon)
  • Dynamic manipulation of history during task execution
  • Built-in caching with Helicone if enabled.

Installation

To get started with BeeBot, you can clone the repo to your local machine and install its dependencies using poetry. These instructions may vary depending on your local development environment.

git clone https://github.com/AutoPackAI/beebot.git
cd beebot
./setup.sh

Windows is officially unsupported but it may work. PRs are welcome for Windows compatibility but will not be a primary focus.

Persistence

Persistence is required. While SQLite is officially supported and is used in tests, it is highly recommended that you use Postgres via docker, simply by executing docker compose up -d.

Running

CLI

To use the CLI run:

poetry run beebot

API (via e2b)

To start the server run:

uvicorn beebot.initiator.api:create_app --factory --timeout-keep-alive=300

If you're doing development on BeeBot itself, you may want to use this command:

uvicorn beebot.initiator.api:create_app --factory --reload  --timeout-graceful-shutdown=3 --timeout-keep-alive=300

and then you can call the API using the following commands:

To create a task run:

curl --request POST \
  --url http://localhost:8000/agent/tasks \
  --header 'Content-Type: application/json' \
  --data '{
	"input": "Write '\''hello world'\'' to hi.txt"
}'

You will get a response like this:

{
  "input": "Write 'hello world' to hi.txt",
  "task_id": "103",
  "artifacts": []
}

Then to execute one step of the task copy the task_id you got from the previous request and run:

curl --request POST \
  --url http://localhost:8000/agent/tasks/<task-id>/steps

Websocket Connection

Note: Notifications are currently undergoing a rework and may not work at the moment

To receive a stream of changes to all the data models in BeeBot, you can subscribe to the websocket connection at the /notifications endpoint with the same host/port as the web api, e.g. ws://localhost:8000/notifications. Use your favorite websocket testing tool to try it out. (I like Insomnia)

Web Interface

We are working on a web interface using Node.js (Remix)

Philosophy

BeeBot's development process is guided by a specific philosophy, emphasizing key principles that shape its development and future direction.

Priorities

The development of BeeBot is driven by the following priorities, always in this order:

  1. Functionality: BeeBot aims to achieve a high success rate for tasks within its range of expected capabilities.
  2. Flexibility: BeeBot strives to be adaptable to a wide range of tasks, expanding that range over time.
  3. Reliability: BeeBot focuses on reliably completing known tasks with predictability.
  4. Efficiency: BeeBot aims to execute tasks with minimal steps, optimizing both time and resource usage.
  5. Convenience: BeeBot aims to provide a user-friendly platform for task automation.

Principles

To achieve these priorities, BeeBot follows the following principles:

  • Tool-focused: BeeBot carefully selects and describes tools, ensuring their reliable use by LLMs. It uses AutoPack as the package manager for its tools.
  • LLM specialization: BeeBot will leverage a variety of LLMs best suited for different tasks, while OpenAI remains the primary LLM for planning and decision-making.
  • Functionality and flexibility first: BeeBot prioritizes functionality and flexibility over developer quality-of-life, which may limit support for specific platforms and other deployment conveniences.
  • Unorthodox methodologies: BeeBot employs unconventional development approaches to increase development speed, such as the absence of unit tests. Instead, end-to-end tests are used, ensuring the entire system works together as expected.
  • Proven concepts: BeeBot adopts new concepts only after they have been proven to enhance its five priorities. As a result, it does not have complex memory or a tree of thought.

Documentation

For further information on the architecture and future plans of BeeBot, please refer to the docs/ directory. The documentation is currently very light, but will evolve alongside the project as new insights and developments emerge. Contributions and feedback from the community are highly appreciated.