AI Code Review for Startup CTOs & Founders
Argus reviews PRs on your own LLM key and keeps what your team teaches it in memory after the people who taught it move on. There is no paid tier.
Last updated
What problems do Startup CTOs & Founders face in code review?
- Small teams can't afford dedicated reviewers — the CTO reviews everything, and it's a bottleneck
- Key engineer leaves and takes all the architectural knowledge with them
- Moving fast means cutting corners, and tech debt compounds faster than you can track it
- Junior engineers merge code that senior engineers would have flagged — but seniors are busy shipping
How does Argus fit your workflow?
- With SELF_HOSTED=true, auto-review is on by default and Argus reviews each PR when it opens. With SELF_HOSTED unset or false it defaults to off, and a repo or org setting can turn it on or off under either default. When it is off, no review runs until someone with write access ticks the 'Trigger Argus review' checkbox or comments @argus-eye review (the handle is your GitHub App's slug; argus-eye is the default)
- Conventions Argus reads from diffs and patterns taught with @argus-eye remember stay in the repo's review memory, and rules added in the dashboard apply to every repo in the installation
- By default each non-skipped file gets one review call under a written 12-rule rubric, and files on auth, token, session, and similar paths get a security-focused pass. Deep Review, off by default, adds four specialist passes on files triaged for full review
- A 👎 (read when that PR is next reviewed or replied to) or a reply explaining why a finding is wrong, from someone with write access, becomes a dismissal. Similar findings in later reviews of that repo are then dropped or downgraded; security findings are only downgraded
Which Argus features matter most for your team?
- Review Laws rubric
- Every review call gets the same 12 written rules in its system prompt, covering what counts as blocking, what evidence a finding needs, and what is out of scope; style and naming are never findings. Deep Review adds bug-hunter, security, architecture, and regression passes when you turn it on
- Review memory
- Patterns people teach, rules they add, and findings they dismiss stay in memory and feed later reviews after those people leave
- Convention learning
- Each review extracts up to 3 conventions from the added lines. When a new one contradicts a stored one, Argus posts a PR comment asking the team which one stands
- PR description diagrams
- Up to two Mermaid diagrams are added to the PR description, drawn only from code-graph edges between the changed files. A diagram with an edge the graph does not contain is dropped
When auto-review is off (the default unless SELF_HOSTED=true), each new PR gets a checkbox comment listing files and diff lines and, once the repo has completed reviews, the average tokens (and cost, when recorded) of its last 20; no review runs until someone with write access ticks the box or comments @argus-eye review