Getting Started
Revet is a developer-first code review agent that combines deterministic static analysis with selective LLM reasoning. It builds a persistent code intelligence graph first, then uses AI only for ambiguous findings.
Install
cargo install revet
Initialize
Run this once in your project root to create a .revet.toml config file:
revet init
First run
Review changes against your main branch:
revet review
Review the entire codebase:
revet review --full .
What happens
- File discovery — finds all supported source files (diff-based by default,
--fullfor everything) - Code graph — tree-sitter parses each file into a dependency graph (incremental: unchanged files load from cache)
- Baseline graph — loads the previous run's graph to detect breaking changes
- Impact analysis — detects breaking changes affecting other parts of the codebase
- Domain analyzers — regex-based pattern scanning for security, ML, infra, and more
- Suppression — filters out inline-suppressed, per-path-suppressed, and baselined findings
- Output — findings printed to terminal (or JSON/SARIF/GitHub annotations)
- Run log — full results (kept + suppressed) written to
.revet-cache/runs/<id>.json
Viewing suppressed findings
By default, suppressed findings are silently filtered. Pass --show-suppressed to see them with their reason:
revet review --show-suppressed
Inspecting past runs
Every review writes a detailed log. View it with:
revet log # list recent runs
revet log --show <id> # full JSON for a specific run
The terminal summary always shows the exact command at the end:
Run log: revet log --show 1772142454966