Skip to main content

Frequently Asked Questions

Q: Where can I ask questions?

A: Try the Nopilot Discord, #questions channel.

Q: What is the difference between an Agent an Assistant?

A: See Terms below.


Assistant: A program that helps a human with a task by taking roughly 1 action in response to a user request.

For instance, GitHub Copilot, or

Agent: A program that can take many actions in response to a single user request, changing course due to environment feedback.

For instance, OpenDevin, SWE-agent, or Github Copilot Workspace.

Autonomous DevTool: An agent that is also a DevTool, suitable for work on production systems.

SWE-bench: The defacto standard metric Coding Agent performance on realistic tasks. Cases are taken from real GitHub issues on popular Python repositories. Many notable groups have submitted scores inclding Amazon, IBM, Cognition (Devin), and OpenDevin. "SWE" is short for Software Engineering.

SWE-bench-lite: The official subset of SWE-bench examples, created because the full benchmark is costly to run.

Open Source: Licensed to allow all of the folowing: use for any purpose, modification of the source, distributing the original, distributing modified copies.

Source Available: Source can be publicly inspected but not neccesarily allowing Open Source use.

Open Weight: Neural Network weights released under a permissive license. Most are not "Open Source" because weights are not "source", the training data and process are.

ReAct: Short for Reason Act, this prompting technique enables LLM problem solving with tool-use (not to be confused with the React Javascript framework). Several coding agents are based on it, including SWE-agent and CodeAct.