NVIDIA’s Boro is an AI-assisted command-line tool for Linux kernel patch development in a developer’s own Git tree. It supports patch review, builds, targeted tests, and AI-assisted patch application—work aimed at local development and downstream maintenance, such as backports and distribution kernels.

Andrea Righi presented Boro during the Linux Plumbers Conference 2026, held in Prague from October 5 to 7. The conference listing raised whether some Boro functionality could be integrated into Sashiko, an AI-assisted patch-review tool.

What Boro does in a local Git tree

Boro’s review, build, test, and apply commands address different stages of kernel patch work. They operate on commits or commit ranges in a developer’s tree, helping with review and validation before patches are posted publicly or as part of downstream maintenance.

The review command runs a multi-stage, kernel-focused review of a commit range and produces a narrative summary in the style used for Linux kernel mailing list discussions. Its pipeline numbers stages from 0 through 12, moving from identifying relevant subsystem guidance and summarizing upstream follow-up to reviewing execution flow, resource lifetimes, locking, security, and portability. Later stages consolidate findings and validate additions.

From review to builds and targeted tests

The build command checks out commits into separate Git worktrees, then builds them and sends the build output for model-assisted triage. Separate worktrees let the workflow handle commits independently while keeping the developer’s main working tree distinct.

The test command can boot a kernel in a virtual machine using virtme-ng and run a targeted test selected by a model. The project describes matching kernel self-tests, userspace probes of changed code paths, and dmesg as a fallback. Hardware-dependent or longer-running tests can be described in a plan without being run. The test workflow requires virtme-ng to be installed.

For patch application, boro apply can apply commits or a mailing-list series fetched with b4. When a cherry-pick leaves certain diff3 conflicts, Boro can propose resolutions and have a model validate them. That makes conflict handling part of an assisted workflow, rather than a claim that proposed changes are automatically accepted.

Model options and dependencies

The NVIDIA/boro project documents configurable OpenAI-compatible API endpoints through BORO_URL, BORO_KEY, and BORO_MODEL. It also lists claude, opencode, and codex as agent-backend options. These are ways to configure Boro to work with agent command-line tools, not software bundled with the project.

The required utilities depend on the task: virtme-ng supports the build and test workflow, while b4 is needed to apply a mailing-list series by message ID. lei is an optional tool for retrieving messages from lore.kernel.org.

Boro and Sashiko

The Linux Plumbers Conference talk description framed possible integration of some Boro functionality into Sashiko as a discussion topic. It did not describe a completed integration.

Language and license

Boro is written in Rust and licensed under Apache 2.0.