llmvet is a local code-review tool for LLM/agent harnesses. It shows a diff in the browser, lets a human reviewer leave inline comments, and then prints a prompt (with the comments) to stdout for the agent to consume.
When the reviewer approves without changes, llmvet exits with code 0 and no output. When they submit comments, the prompt is written to stdout — the agent can capture it from there.
An agent calls llmvet after making changes to the working tree:
bin/llmvet
This starts a local web server, opens a browser tab showing the diff, and
blocks until the reviewer either submits comments or approves. When the
reviewer submits, the review prompt (one > -blockquoted comment per inline
note) is printed to stdout, which the agent can capture and act on.
You can run the review yourself from the command line. The output is the same prompt format that an agent would see — copy it into your LLM context:
bin/llmvet
# opens http://127.0.0.1:<port>/ in your browser
# review the diff, leave comments, click Submit
# the prompt with comments is printed to stdout
Options:
-port <n>— bind on a specific TCP port (default: random free port)-no-open— print the URL instead of opening the browser automatically
| Code | Meaning |
|---|---|
| 0 | Approved (no output) or submitted (prompt on stdout) |
| 130 | Review aborted (Ctrl-C, SIGTERM, browser closed without acting) |
make web # build the frontend bundle (writes internal/assets/dist/)
make build # compile bin/llmvet with the bundle embedded
make test # run Go tests
make clean # remove build artifacts
make build runs make web first because //go:embed requires the vite
output to exist at compile time.