GitHub - gi-dellav/rv: Non-invasive AI code review for any type of workflow.

rv

rv is a non-invasive AI code review for any type of workflow.

It works as a CLI tool easy to use and integrate, allowing to review the code that you are currently writing or code from other commits, branches or pull requests.

Features

Unix philosophy
rv follows the Unix philosophy by providing a minimalstic tool (~1.5k LoC) that does one thing well.
Cheap and low-latency
rv is optimized to use cheap and low-latency models in order to allow for reviews that takes less than 10 seconds and cost about $0.002 (on average, tested with Qwen3-32B)
Fully customizable
rv is designed to give full freedom with its configuration file, allowing for different providers, LLMs and prompts
Semplicity of code
rv is designed to be written using clean, understandable and safe (as in no unsafe instructions used) Rust code
Open source and non-monetized
rv is released under the GPL license and we will never sell subscriptions, cloud credits or other form of monetized services to our end users

How To Install

From crates.io

Just run cargo install rv-tool in order to install the last version (the specified version is only needed on testing releases); then, follow the "From the source" guide from the third step.

From the source

Clone the repository and compile using cargo install --path .

Finish configuration

Run for the first time (just rv) in order to generate the configuration file
Edit the ~/.config/rv/config.toml file setting up provider, model and API key (if you don't want to use ENV variables)
rv is now installed and ready! Run rv while you have staged edits (aka after git add) in order to get a code review of your current progress

NOTE: rv has been only tested on Linux; if possible try it on MacOS and Windows and open an issue with the results.

How to setup APIs

We reccomend using OpenRouter as it allows to use different models, connect to different APIs (such as Azure, Anthropic, Cloudflare, Google and Mistral), and access some free models. Here are the links for creating an account, managing API keys, connecting other provider and viewing all free models. Once you have the API key, you can insert it in your configuration file (on Linux, ~/.config/rv/config.toml).

How to use

For reviewing staged changes or the last commit: rv

For reviewing a specific commit: rv -c [commit]

For reviewing a specific branch: rv -b [branch]

For reviewing a Github PR: rv -p [pr-id] (Requires gh to be installed)

For switching to a different LLM profile: rv -l [llm]

For reviewing files without the Git integration: rv --raw

NOTE: If you want to use the output for shell pipes or for writing to a file, use the -P/--pipe flag.

Model profiles

The current suggested models is qwen/qwen3-235b-a22b-2507 (for the default profile) and deepseek/deepseek-v3.2 (for the think profile) for more intensive tasks. You can switch between different profiles using the -l CLI flag and you can add or remove profiles from ~/.config/rv/config.toml.

Future work

Milestones planned for the future:

make Action Menu configurable
allow for referencing files outside of the current review (full project context)
support for larger reviews by splitting into file-by-file review (requires full project context implementation)
ability to add context sources from the chat tool or from project files
ability to use regex rules (with $any[], $all[] and $none[]) inside of project files and custom prompts
ability to load PDF files and images as context sources (useful for documentation, specifications, etc)
fix tool for producing and applying fixes directly from the review
integration with git hooks
integration with language-specific CLI tools (ex. cargo check)
support for MCP servers
text mode for reviewing content and style of natural language documents (ex. essays), with support for TXT, MarkDown, LaTex.
markdown rendering with external tools (ex. glow)
support for other local and cloud LLM providers

rv