System requirements
Application and database
A Linux server with Docker, 2 to 4 CPU cores, 8 GB of memory and enough disk space for the files and backups - disclosures, drafts and office letters take little space, scanned documents and drawings more. The web app runs in current versions of Chrome, Edge, Firefox and Safari.
AI model
The included model (Qwen 2.5 14B) needs about 10 GB of memory for itself. On a server with a suitable GPU (16 GB of video memory or more) a complete draft usually takes a few minutes. On a CPU-only server it works as well, but a draft takes 25 to 75 minutes; drafts run one after the other in the background, while short tasks such as revisions are served between the steps of a running draft.
Whether AI and database share a server or the AI gets its own is decided at installation; the AI server only needs to be reachable by the application. Without a local model (external providers only) or without any AI, there is no AI server; the server for application and database is enough.
Backups
Backups should go to another disk or a network drive; they are encrypted and can also be copied off the server.
Connecting AI assistants (MCP)
The Model Context Protocol (MCP) connects AI assistants such as Claude to PatentFlow. The assistant can search clients and files, read files, drafts and claims, list deadlines, record time entries and create tasks. The interface calls the same routes as the web app, so rights, visibility and the audit log apply unchanged, and every tool call is logged.
Three gates protect the connection: the firm switches it on as a whole, each person signs in with their own access, and each client must be released for it. Sign-in uses OAuth 2.1 with a consent page in PatentFlow; clients without OAuth can use personal tokens. The server must be reachable over HTTPS for this. The provider of the assistant processes the answers; the release per client controls which data can reach it.