Install
pip install flue && flue setupWho it suits
If you want a programming assistant to drive desktop software from the terminal—the page names Photoshop, Premiere, Blender, Unity, and more—Flue is a fit. It is a very small bridge: the assistant sends a script into the application's own automation runtime through a bridge command, and the application returns JSON.
The assistant environments named on the page include Codex, Claude Code, Gemini CLI, Antigravity CLI, OpenCode, and similar tools. These are examples, not a closed list. As long as your assistant can run a Python package and accept flue setup to discover the current environment, you can make a first attempt.
It is best suited to people who already have at least one of the desktop applications above installed locally but, on the first try, only want to confirm that the assistant can see Flue. Sending a script into a running Photoshop or Blender will change your project, so stop at installation and detection the first time.
If you do not have pip yet or do not want setup to change your assistant configuration, do not start yet.
Complete the first try in 3 minutes
Use only this official installation path, with the two steps together; do not split them into another installer:
pip install flue
flue setup
pip install flue writes the package into the current Python environment, while flue setup detects your assistant environments and lets them know Flue is available. Both steps change local machine state. Before running them, confirm that this is the Python environment and these assistants you intend to use, and that you accept where the changes will be written.
Prerequisite: pip must already run on the machine. Installation goes into the current Python environment and the assistant configurations detected by setup.
After both steps finish, make one low-risk confirmation: open Codex, Claude Code, or another detected assistant and check whether it already knows that Flue is available. The success signal is that the assistant acknowledges it can use the Flue bridge, while you have not yet sent a script to any desktop application.
The page does not specify the bridge executable's name, a script example, or the JSON fields. Do not add a Photoshop or Blender script on the first try. After confirming that the assistant can see Flue, decide separately whether to send a script into an already open application.
Use cases
Case one: confirm after installation that the assistant knows Flue. The page states that flue setup detects your assistant environments and lets them know Flue is available. After running the two official steps, ask about Flue's availability in one detected assistant. The current instructions do not provide a detection list or success wording; use the assistant's clear confirmation of availability as the criterion.
Case two: send a script through the bridge into an application and get JSON back. The page states that the assistant sends a script, the bridge executes it in the application's own automation runtime, and the application returns JSON. Take this step only when the target application is already open and you accept that the script will change it. The current instructions do not provide the script contents or response fields; start with the shortest, reversible read-only intent, and confirm once more before writing the script.
Boundaries
The official installation is only pip install flue followed by flue setup. On the first try, only confirm that the assistant knows Flue. The named desktop applications and assistants are examples. The bridge executes scripts in the application's automation runtime and returns JSON; this step changes the project currently open, so confirm the target application first. The page does not provide a script template or command-line switches; do not add them to the first attempt.