The PSIPRED Protein Analysis Workbench, a web server offering protein prediction tools, has been in operation for 20 years. The authors describe updates to the workbench to prepare it for the next 20 years. The main focus of the upgrade was to improve analysis speed due to the growing size of protein sequence databases. New software, hardware, and web services were introduced. Key predictive algorithms were also updated. The website was redesigned to make predictive methods more accessible and results more intuitive. Modern web frameworks were used, and server processing was moved to the user's browser to reduce backend load. The new site allows users to select input data type (protein sequence or PDB) and presents appropriate methods. Results are now presented in five sections, including a title banner, download panel, segment resubmission widget, sequence annotation plot, and additional results panels. New methods include PSIPRED 4, FFPRED 3, MetaPSICOV 2, DISOPRED 3, DomSerf 2.1, and BioSerf 2. These methods have been improved for accuracy and performance. The backend server was upgraded with more powerful hardware, reducing queue times and runtimes. Sequence search and results caching were improved using HHBlits, which is faster than PSIBLAST. A workflow management engine was developed to streamline the addition of new predictive methods. The PSIPRED Workbench now supports a Service Oriented Architecture, allowing all methods to be accessed via REST. The server is now ready for the next 20 years of bioinformatics progress. The workbench is available at http://bioinf.cs.ucl.ac.uk/psipred.The PSIPRED Protein Analysis Workbench, a web server offering protein prediction tools, has been in operation for 20 years. The authors describe updates to the workbench to prepare it for the next 20 years. The main focus of the upgrade was to improve analysis speed due to the growing size of protein sequence databases. New software, hardware, and web services were introduced. Key predictive algorithms were also updated. The website was redesigned to make predictive methods more accessible and results more intuitive. Modern web frameworks were used, and server processing was moved to the user's browser to reduce backend load. The new site allows users to select input data type (protein sequence or PDB) and presents appropriate methods. Results are now presented in five sections, including a title banner, download panel, segment resubmission widget, sequence annotation plot, and additional results panels. New methods include PSIPRED 4, FFPRED 3, MetaPSICOV 2, DISOPRED 3, DomSerf 2.1, and BioSerf 2. These methods have been improved for accuracy and performance. The backend server was upgraded with more powerful hardware, reducing queue times and runtimes. Sequence search and results caching were improved using HHBlits, which is faster than PSIBLAST. A workflow management engine was developed to streamline the addition of new predictive methods. The PSIPRED Workbench now supports a Service Oriented Architecture, allowing all methods to be accessed via REST. The server is now ready for the next 20 years of bioinformatics progress. The workbench is available at http://bioinf.cs.ucl.ac.uk/psipred.