BESSER Bot Framework

This BESSER Bot Framework review covers what the framework is used for, its key features, what it costs, how the rename to BESSER Agentic Framework affects you, where it shines and falls short, and how it compares with other tools. The BESSER Bot Framework (BBF) is part of BESSER (Building Better Smart Software Faster), an open source low-code project at the Luxembourg Institute of Science and Technology (LIST), which aims to help teams build better smart software faster. In February 2025 the project was renamed the BESSER Agentic Framework (BAF) to reflect its shift from rule-based bots to AI agents (LIST announcement). The old besser-bot-framework package on PyPI stops at version 1.5.0; new work ships as besser-agentic-framework, version 4.5.2 at the time of writing. You write an agent in plain Python code: states, transitions between them, and the intents that trigger those transitions.

What is BESSER Bot Framework used for?

The framework is used to build chatbots and AI agents that follow a clear conversation flow but can still understand free text. Typical examples are customer support and FAQ bots, assistants that answer questions from your own documents with RAG, bots that react to events in a GitHub or GitLab repository, Telegram bots, and multi-agent systems where one agent talks to the user and others generate or review code. Researchers use it to experiment with agent design, and teams use it when they want an agent they can run on their own servers instead of a hosted chatbot platform.

Because an agent is a finite-state machine, you can see and test every step of a conversation, which suits deterministic workflows such as support flows or form-like dialogues. LLMs handle the parts where rules are not enough: classifying intents, extracting data from user input or writing a reply. The result is a hybrid system that combines structured logic with generative capabilities. Being pure Python also makes it easy to integrate data science libraries, databases and APIs into an agent.

Key features

• State machines: an agent is a set of states, each with a body (Python code that runs on entry) and transitions triggered by intents, events or conditions.
• Intents and entities: define intents with example sentences and extract entities from user input with named entity recognition. The simple intent classifier has a PyTorch implementation and a TensorFlow implementation, and an LLM-based classifier is available.
• LLMs: connect OpenAI, HuggingFace models, Replicate, Groq, Mistral and others, and mix LLM replies with rule-based logic.
• RAG: answer from your own documents with Retrieval Augmented Generation.
• Platforms: a WebSocket platform with a Streamlit UI or chat widget, plus Telegram, GitHub, GitLab and an A2A platform for agent-to-agent communication.
• Speech: speech-to-text and text-to-speech for voice agents.
• Multi-agent systems: since version 2.0 agents can talk to each other and split a task between them.
• Monitoring: a database and dashboard that track sessions and agent interactions, so you can check the details of past conversations.
• Languages: agents can run in several languages; the agent's language set to Luxembourgish needs the spellux library, which you manually install.

BESSER Bot Framework pricing

The framework is free. It is open source under the MIT license, so you can use it in commercial projects, change the source code and self-host it without a license fee (GitHub repository). There is no paid plan, no hosted version and no usage limit from the project itself.

The cost to check is everything around it. LLM providers such as OpenAI charge per token, hosting the agent and its WebSocket server is your own bill, and optional dependencies like PyTorch or TensorFlow need a server with enough memory. Support comes from the GitHub issues and the documentation, not from a vendor contract.

Installation and example agents

The framework needs Python 3.10 or later. Create a virtual environment, then install the base package with pip:

pip install besser-agentic-framework

This command installs the base package with the core dependencies only. To add the necessary dependencies for additional agent functionalities, use the following tags: extras (RAG, speech to text, plotly, opencv), llms (the additional dependencies to run LLMs: OpenAI, Replicate, transformers), torch to install PyTorch for the simple intent classifier and HuggingFace models, tensorflow (a very heavy package, so install it only if necessary), docs to compile the project documentation, and all for all the dependencies at once. If you cloned the repository, install from the requirements files instead. The documentation has details on every option and a running example for each concept. The repository has example agents to start with: a very simple agent for the first contact (greetings), a weather agent with entities, an LLM agent, a RAG agent, Telegram, GitHub and GitLab agents, and an A2A multi-agent example. More live in the BAF-agent-examples repository.

A practical order: create the virtual environment, run the install command for the base package, and start with the very simple agent before you add any heavy package. Add the torch or tensorflow tag only when you need the simple intent classifier on your own server; its PyTorch implementation and TensorFlow implementation need the same necessary dependencies as HuggingFace models, and the project itself calls TensorFlow a very heavy package. If the agent's language is Luxembourgish, manually install spellux as well. Keeping the core dependencies small makes the system easier to deploy in a container and to integrate with an existing Python backend.

From BESSER Bot Framework to the open source BESSER Agentic Framework

BESSER Bot Framework and BESSER Agentic Framework are the same project. The rename came with version 2.0, which added multi-agent support, and the documentation, repository and package now use the new name (documentation). Old links to the BESSER Bot Framework docs no longer resolve, and the old GitHub repository redirects to the new one. Projects on BBF 1.x keep working on the old package, but new features only land in BAF, so start new projects on the new package. The framework also relates to the wider BESSER low-code platform, whose foundation is the B-UML modeling language. The research team presents the framework's state machines as a proving ground for how B-UML will model smart software, and BESSER runs as a five-year project funded by the Luxembourg National Research Fund (arXiv paper).

Where BESSER Bot Framework shines — and where it falls short

Strengths:

• Free and open source under MIT, with no vendor lock-in.
• Clear state-machine model that makes conversations testable.
• Mixes rule-based flows with LLMs, RAG, speech and multilingual NLP in one framework.
• Active development by a research institute, with regular releases.

Trade-offs:

• Code only: there is no visual builder or drag-and-drop interface for non-developers.
• State machines make free-flowing, open-ended conversations harder to design.
• A small community compared with the large agent frameworks.
• Heavy optional dependencies for local models.
• Documentation and examples assume you know Python.

How BESSER Bot Framework compares

• BESSER: the low-code modeling platform the framework belongs to.
• Voiceflow: a hosted visual builder for chat and voice agents, aimed at product teams.
• Landbot: no-code chatbots for websites and WhatsApp, with a monthly subscription.
• Chatfuel: no-code bots for Facebook Messenger, Instagram and WhatsApp.
• n8n: workflow automation with AI agent nodes, self-hostable like BAF.
• AI Builder: Microsoft's AI models for Power Apps and Power Automate.

One third-party alternatives site (NoCodes.ai) describes the framework as a no-code chatbot builder and warns about on-premise limits; the project's own documentation shows a Python library you install and host yourself, so neither claim holds.

Is BESSER Bot Framework worth it?

The BESSER Bot Framework, now the BESSER Agentic Framework, is worth it for Python developers who want a free, self-hosted framework for chatbots and AI agents with a clear structure, and for teams that need to mix fixed conversation flows with LLMs. It is a good fit for research, internal assistants and developer tools on GitHub or GitLab. It is not the right choice for non-developers who need a visual builder, or for teams that want a hosted platform with support and integrations out of the box; Voiceflow or Landbot suit them better. If you want help to design or build agents with it, the agencies and developers on LowCodeDevs can take it on.

Sources

• BESSER Agentic Framework on GitHub: README, installation, license and example agents.
• BESSER Agentic Framework documentation: core concepts, platforms, NLP and LLM support.
• From Bots to Agents, LIST: the February 2025 rename and version 2.0.
• Building BESSER: an open-source low-code platform, arXiv: the framework's role in BESSER and FNR PEARL funding.
• BESSER Bot Framework alternatives, NoCodes.ai: third-party alternatives list (its description conflicts with the project's documentation).