ChimeraX OpenFold Prediction Server

Tom Goddard
February 11, 2026

ChimeraX can run OpenFold predictions on a server that you set up. For example, ChimeraX prediction jobs can be sent from a Mac or Windows laptop to a Linux computer with Nvidia graphics to make faster predictions of larger molecules.

Server setup

To setup a server (can be Linux, Windows, or Mac) you install ChimeraX (February 13, 2026 version or newer) on the server computer, then run the following ChimeraX commands to install openfold and start the server process. The setup commands can be run without a graphical display (for example, if the server has no display) by starting ChimeraX in text only mode.

      chimerax --nogui
      > openfold install
      > openfold server start
      > exit

When you quit ChimeraX the server remains running. The server listens for prediction requests from ChimeraX from other computers, runs the predictions, and returns the results. By default the server keeps temporary job files in directory ~/openfold_server_jobs. Multiple predictions can be queued but only one runs at a time, unless multiple GPUs are available and the gpus option is used when starting the server. The OpenFold server can be setup on cloud virtual machines from commercial providers.

How to run a prediction on the server

OpenFold predictions are run from ChimeraX in the usual way, only the server host name is specified. Use ChimeraX menu Tools / Structure Prediction / OpenFold to specify the molecules to prediction. Press the Options button on the OpenFold panel, check User server and provide the host name of the server. The press the OpenFold Predict button to run the prediction. To always use the server click the Save default options button.

The predicted structure will automatically be opened when the calculation finishes. You can quit ChimeraX and later restart it and fetch new predictions from the server using ChimeraX menu Tools / Structure Prediction / OpenFold History. If there are active server predictions the OpenFold History panel will have a button Fetch from server that will get any completed predictions. Note locally run predictions behave differently and are terminated if you quit ChimeraX.

Result file location

To use the prediction server ChimeraX copies the folder with the OpenFold input (.json file, msas, command options) to the server machine, OpenFold runs on the server, and the results are copied back to the machine that made the request. All OpenFold input and output will be in a folder the same as if the prediction were run on the local machine.

Server commands

The command to start the openfold server has the following options.

      openfold server start [host hostname] [port port-number] [openfoldExe executable-path]
                         [jobsDirectory directory] [device gpu | cpu] [gpus gpu-ids] [extraOptions openfold-options]

Options descriptions

host hostname - The host name of the server, for instance minsky.cgl.ucsf.edu or 169.230.21.20. The default uses the python gethostbyname() system call to get the host name or IP address.
port port-number - The port number to listen for connections. Default 30173.
openfoldExe executable-path - The path to the openfold executable. The default uses the OpenFold install location which is initially set by the ChimeraX openfold install command and can later be changed using the ChimeraX OpenFold Options panel or with the openfold predict command installLocation option.
jobsDirectory directory - Directory where server job subdirectories will be created. The default is ~/openfold_server_jobs. The directory will be created if it does not exist.
device gpu | cpu - Whether the server should run OpenFold using the GPU or the CPU. Default is to run with the GPU.
gpus gpu-ids - Comma-separated list of integer GPU ids. This allows the server to run multiple prediction jobs at the same time, one per GPU. By default only one GPU is used even if the computer has multiple GPUs.
extraOptions options - Additional options to add to the openfold command-line for every prediction. For instance, low memory OpenFold options "--use_cpu_memory --inplace_operations" can allow predicting larger structures.

A server log text file named "openfold_server_log" is created in the jobsDirectory when the server is started and can be used to debug server errors.

List and fetch active server jobs

If you quit ChimeraX and later restart it it remembers server jobs you were running that were not completed. The OpenFold History panel (menu Tools / Structure Prediction) will have a button Fetch from server that can download and display completed server jobs.

Alternatively you can list these "active" jobs with the openfold server list command and fetch the resulting predictions with the openfold server fetch command. When results are fetched the structures will be opened in ChimeraX. If pending jobs are not complete the fetch command will report the current state (running, waiting to run).

      openfold server list
         - Lists active server job identifier and server host and port.

      openfold server fetch [local-directory] [open true | false]
         - fetch server predictions

Options descriptions

local-directory - The openfold job directory on the local computer. If this is not specified then all "active" openfold server jobs are fetched. Active jobs are ones that were started in ChimeraX and did not complete before ChimeraX was exited. Output to the Log will indicate for each pending job whether it has finished, running, or waiting to run.
open true | false - Whether to open the prediction structures in ChimeraX. Default true. If you do not open them, they can later be opened using the ChimeraX OpenFold History panel from menu Tools / Structure Prediction / OpenFold History.

Stopping the OpenFold prediction server

There is currently no command to stop the OpenFold server process. On Linux or Mac you can determine the process id and kill the process using shell commands.

      $ ps -axww | grep openfold
        34435 ChimeraX.app/bin/python3.11 ChimeraX.app/lib/python3.11/site-packages/chimerax/openfold/server.py ...
      $ kill 34435

On Windows use Task Manager to find the openfold server process and force quit it. On Mac you can also use Activity Monitor to find the openfold server process and force quit it.