Patched

A patchflow is a ready-made automation that combines reusable steps and prompt templates into a job that runs end to end. An AI patchflow handles the full chore, not just a single call. Out of the box there's PR review, AutoFix for security vulnerabilities, DependencyUpgrade, ResolveIssue, GenerateReadme, and GenerateDocstring, each one built to take a repetitive task off your plate and carry it through to a finished result without you babysitting every step. You can chain your own if the built-in set doesn't fit.

Patchwork runs as a CLI agent on your own machine or server, so your code and data don't have to leave your environment. Install it with pip and point it at your preferred LLM, whether that's OpenAI, Google, or the Patched managed service. This matters when a compliance team needs to see where the data actually goes. It usually does.

The framework isn't tied to one model provider. You can swap in an OpenAI key, a Google model, or the managed Patched key, and even mix providers across patchflows. That flexibility keeps you from being locked into a single vendor's pricing or rate limits. Not a small thing.

Patchflows aren't confined to the terminal. You can wire them into a CI/CD pipeline so reviews, fixes, and upgrades run on every commit, or call them from inside your IDE while you work. The same automation that reviews a PR can patch a vulnerability before it ships. For teams buried in pull requests, that's AI code review automation without hiring another reviewer.

Every workflow is designed to be fully auditable, which is the whole point for regulated industries. Patched deploys systems on dedicated infrastructure and takes ownership of the SLA, so you get a paper trail and an accountable operator rather than a black box. As a workflow execution platform, it owns the outcome, not just the wiring, which is a meaningful shift for any team that has been burned by a tool that automated the easy part and left the hard decisions to a human at 2 a.m.

When the built-in automations fall short, you build your own by combining steps and prompt templates. A step is an atomic action like creating a PR or committing a change, and a prompt template shapes how the LLM handles a specific chore. That's how Patchwork turns a repetitive task into something you run once and forget. You write it once. It runs every time.

You can run Patchwork yourself for full control, or use the hosted version at app.patched.codes if you'd rather not manage infrastructure. The hosted route is the faster on-ramp; the self-hosted route is for teams with strict data-residency rules, where keeping code and credentials inside your own network isn't a preference but a hard requirement written into policy.