The Pulse
NVIDIA Publishes Boro for AI-Assisted Linux Kernel Review
NVIDIA’s Boro is an Apache-2.0 command-line tool for reviewing, building and testing Linux kernel patches. Andrea Righi presented the project at Linux Plumbers Conference on October 6, 2026.
AI.info Team · Written with AI ·
NVIDIA’s Boro brings AI-assisted review, building and testing into the Linux kernel developer’s local workflow. Andrea Righi of NVIDIA presented the tool at Linux Plumbers Conference in Prague on October 6; its public repository describes a command-line program for working with patch stacks before submission or in downstream maintenance.
Boro targets patches before mailing-list review
Boro is designed for developers working inside their own Git trees, including those preparing a patch series, backporting changes or integrating security fixes. That focus distinguishes it from Sashiko, the project that inspired Boro: Sashiko is geared toward reviewing patches already posted to public mailing lists, while Boro handles interactive local review and repair.
The command-line interface accepts a commit range, such as origin/master..HEAD, and reviews each commit through a sequence of model-assisted checks. Its stages cover kernel-subsystem identification, upstream follow-up research, control flow, resource management, concurrency, security, and hardware and portability concerns. A later validation stage evaluates candidate findings before Boro assembles an LKML-style report.
Separate models handle discovery and validation
Boro lets users assign different models to broad review and validation. The BORO_MODEL setting handles regular discovery, while BORO_VALIDATION_MODEL can handle the initial complete-context review, validation of new findings and report rendering. Users can point the tool at an OpenAI-compatible endpoint, including a local Ollama instance, or use the documented Claude, OpenCode and Codex backends.
The workflow also gathers patch metadata and source context locally before sending review requests. That arrangement gives developers a local-model option, but it does not mean every Boro setup keeps inference on the developer’s machine: the repository also documents remote API endpoints. Model choice and endpoint configuration determine where inference runs.
Build and test add checks beyond the diff
Boro’s build and test commands extend the process beyond reading a patch. Build mode checks out commits in separate worktrees and uses virtme-ng to build the kernel; test mode can boot the result in a virtual machine and run a model-selected targeted test. The tool then asks a model to assess the build or test output.
Those commands require virtme-ng, while applying a mailing-list series by message ID requires the b4 utility. An optional lei installation enables Boro to retrieve follow-up discussions from lore.kernel.org. The project is licensed under Apache 2.0, and its installation instructions use Cargo.
AI review still needs a maintainer
Boro’s multi-stage design makes its review process inspectable: the repository documents which checks run, separates initial findings from later additions, and records token use by stage. It also describes a validation mode that preserves an initial review baseline while checking proposed additions, rather than allowing later passes to silently replace that baseline.
Those controls do not establish that a finding is correct or that a patch is safe. The repository calls Boro an “AI-assisted kernel development CLI”, and its findings come from the models a developer configures, working from the code and test context they are given. The project’s concrete scope is local patch iteration, backport work and optional virtual-machine testing, with the developer responsible for deciding what to accept.