sigmund 1.23.2

AI-based chatbot that provides sensible answers based on documentation

Sigmund AI

Copyright 2023-2026 Sebastiaan Mathôt

A Python library and web app for an LLM-based chatbot:

Features:

• Privacy: all messages and uploaded attachments are encrypted so that no-one can listen in on your conversation
• Expert knowledge: access to documentation
• Continuous conversation: conversations are summarized when they become too long to fit into the prompt
• Tool use (in research-assistant mode):
  • Code execution: ability to execute Python and R code
  • Literature search: ability to search for articles on OpenAlex and Google Scholar
  • Image generation: ability to generate images with Flux.2 and Dall-E 3
• Integrations
  • Python: connect to JupyterLab, Notebook, Spyder or Rapunzel
  • OpenSesame: directly integrate with OpenSesame
  • Sigmund Analyst: directly integrate with Sigmund Analyst for data analysis

Sigmund is not a large language model itself. Rather it uses third-party models. Currently, models from OpenAI, Anthropic, and Mistral are supported. API keys from these respective providers are required.

output2.webm

Configuration

See sigmund/config.py for configuration instructions.

Dependencies

For Python dependencies, see pyproject.toml. In addition to these, pandoc is required for the ability to read attachments, and a local redis server needs to run for persistent data between sessions.

Running (development)

Download the source code, and copy .env.example to .env. Edit this file to specify at least the API keys, and depending on the functionality that you want activate, possibly also other variables. The only variable that is strictly required is the OpenAI API key, because OpenAI is used to create text embeddings, even when a different model is used for the conversation.

Next, install the dependencies, build the documentation index, and launch the app!

pip install .               # install dependencies
python index_library.py     # build library (documentation) index
python app.py               # start the app

Next, access the app (by default) through:

http://127.0.0.1:5000/

Running (production)

In production, the server is generally not run by directly calling the app. There are many ways to run a Flask app in production. One way is to use gunicorn to start the app, and then use an nginx web server as a proxy that reroutes requests to the app. When taking this route, make sure to set up nginx with a large client_max_body_size (to allow attachment uploading) and disable proxy_cache and proxy_buffering (to allow status messages to be streamed while Sigmund is answering).

License

Sigmund is distributed under the terms of the GNU General Public License 3. The full license should be included in the file COPYING, or can be obtained from:

• http://www.gnu.org/licenses/gpl.txt