AI code reviews your agents can drive.
FriendlyReviewer reviews your PRs and MRs with the context of the ticket, your architecture, and your Knowledge Base. No subscription, no seat pricing, no persistent source code storage.
Try this prompt with your AI
I'm considering using FriendlyReviewer for AI code reviews.
Read the complete product reference at https://friendlyreviewer.fr/get/friendlyreviewer.md.
Evaluate whether it fits my team. If you need details about my current review workflow, tools, or constraints to give a useful answer — ask me.
Pricing that does not require a budget committee.
No subscription. No seats. No artificial limit on the developers who can trigger reviews.
$0.50
per analyzed PR or MR. Correction validations are included in the price.
$1
to create the repository's synthetic Knowledge Base. It avoids rediscovering everything on every review.
Refunded
if the review fails technically. Cost should follow value, not incidents.
How it works, no smoke and mirrors.
Yes, we explain part of the recipe. Rebuilding it properly is still another story.
Diff + ticket
FriendlyReviewer fetches the PR/MR, the Jira or Linear context, then focuses on what actually changed.
Knowledge Base
The repo is summarized into concise memory: architecture, conventions, sensitive areas. No need to reread the whole project every time.
Parallelization
The review is split into logical tasks. Multiple agents analyze them in parallel, then an aggregator deduplicates and prioritizes the findings.
Useful comments
Feedback is posted inline, with a summary. The goal: less noise, more actionable points.
Targeted validation
After fixes, FriendlyReviewer rereads the relevant points. No need to pay for a full review again to check three changes.
Up to 100 files
Large PRs happen. We prefer to handle them explicitly rather than pretend they are always a good idea.
GitHub and GitLab
Both platforms are fully supported, with comments posted in the tool where your teams already work.
Jira and Linear
The ticket is not decoration: it is used as the source of intent to verify that the code solves the right problem.
Not magic. Just very guided.
The cost mostly comes from tokens and reasoning time. FriendlyReviewer reduces both with a tightly constrained approach.
Optimized prompts
Agents do not chat freely: they follow strict roles, formats, and review criteria.
Less rediscovery
The Knowledge Base avoids rediscovering the repository architecture on every PR/MR.
Low-cost models
Less expensive models remain effective when they are well guided and fed with the right context.
Your code passes through. It does not move in.
FriendlyReviewer is designed to minimize what is known, stored, and billed. Easier to explain to IT leadership, healthier for everyone.
Built for vibe coding tools.
The reviewer is not just a commenting bot. It can be driven by the AI that is writing the code.
Frugal option: custom binary
A workflow with `fr-agent` can start a review, wait for results, then launch validation from the development environment.
Install fr-agentSimpler option: MCP
The IDE agent calls FriendlyReviewer tools directly: start, wait, validate, close. Less plumbing, more feedback loop.
Set up MCPThe AI drives the loop
It starts the review, reads the comments, applies fixes, then asks FriendlyReviewer to validate that everything is good.
Fast, but not rushed.
On successful internal runs, reviews with comments take 3 min 24 on average. Correction validations take 1 min 09. We advertise under 5 minutes and under 2 minutes to stay conservative.