BeeBot is an autonomous AI assistant, described as a "personal worker bee", designed to autonomously execute a wide range of practical tasks. Project development is currently paused. The author states that current (late 2023) large language models are not yet sufficient for general-purpose autonomous AI tasks, and gives three conditions under which the project might be restarted: LLMs significantly improve in structured thinking, reliable results, and following instructions; a custom model specifically trained for autonomous AI can be developed or fine-tuned; or a specific subset of tasks is found that BeeBot is good enough at and can focus on (the author notes: not programming).
Main capabilities include: tool selection through AutoPack, with the ability to acquire more tools during task execution; built-in persistence; an e2b-compliant REST API; a websocket server for publishing all events within BeeBot; a replaceable filesystem simulation that supports storing files in memory, on disk, or in a database; a Web UI for managing tasks (coming soon); dynamically manipulating history during task execution; and, if enabled, using Helicone to provide built-in caching.
Usage: the repository can be cloned locally and dependencies installed with poetry; these instructions may vary depending on the local development environment. Windows is not officially supported, but it may work; PRs for Windows compatibility are welcome, but it will not be a primary focus. Persistence is required; SQLite is officially supported and used for testing, but using Postgres via Docker with docker compose up -d is strongly recommended. The project provides CLI run commands, server startup commands, development commands, and examples of creating a task and executing a single step of a task through the API (creating a task returns a response containing a task_id, which is used to execute the next step).
Websocket connection: you can subscribe to the /notifications endpoint on the same host/port as the Web API, for example ws://localhost:8000/notifications, to receive a stream of changes for all data models in BeeBot. Note: the notification feature is currently being reworked and may not work temporarily. The author mentions that the Web interface is being developed using Node.js (Remix).
Applicable scenarios: the material does not clearly list specific applicable scenarios; it only states that BeeBot is intended to autonomously execute a variety of practical tasks, and emphasizes principles such as tool focus and LLM specialization.
Limitations: development is paused; Windows is not officially supported; persistence is required; the notification feature may be unavailable; the Web UI has not yet been released; the documentation is currently very brief. The project follows a priority order of functionality, flexibility, reliability, efficiency, and convenience, and adopts principles such as tool focus, LLM specialization, functionality and flexibility first, unconventional methods (such as no unit tests, using end-to-end tests), and only adopting proven concepts. OpenAI remains the primary LLM for planning and decision-making.
