Skip to content

Experiments (Deprecated feature)

An Experiment was the original name for a chatbot in Open Chat Studio. The UI now uses the term Chatbots throughout. You may still see "Experiment" in older parts of the interface, API responses, or documentation.

Historical Notes

An Experiment links all the configuration and data for a chatbot including participant sessions, data, actions etc.

Experiment Types

There were three different types of chatbots that you could build in Open Chat Studio:

  • Base language model
  • Assistant (since removed)
  • Pipeline

Base language model

This kind of chatbot is the most commonly used and simple to configure. It is backed the standard language model APIs such as the OpenAI chat completions API, Anthropic messages API or Google Gemini API.

Chatbots configured this way have all the basic features, such as memory and source material. They can also use advanced features such as Scheduling and Reminders.

Assistant

Removed

Assistant-type chatbots have been removed — OpenAI retired the Assistants API on 26 August 2026. See the migration guide if you still need to move a chatbot off this type.

Assistant chatbots made use of OpenAI Assistants. The main advantage of using Assistants was that your chatbot got access to the OpenAI tools:

Code Interpreter

This allowed the chatbot to write and execute code to accomplish tasks. An LLM node offers the same capability as a builtin tool.

For more information see the OpenAI docs.

This allowed the chatbot to search and reference information provided in uploaded files. Indexed Collections replace it.

For more information see the OpenAI docs.

Pipeline

Pipelines allow you to create more complex chatbots by defining a ‘graph’ (in the computer science sense) of nodes. You can think of this graph as a workflow that flows from input to output. Each message to the chatbot is processed by the graph to produce a final output. A single response from the chatbot will be one successful path through the graph from the input node to the output node.

This can be useful if you want to build a complex chatbot that performs different tasks depending on the participant’s request. A single chatbot prompt that performs several functions generally works poorly. It is better to create one prompt per task and combine them using a Pipeline. This is similar to the Multi-bot setup but allows more flexibility and complexity.

See also