AI code review · GitLab · GitHub
PhonePe Arnold¶
Arnold is a self-hosted code review workspace for GitLab merge requests and GitHub pull requests. It gives reviewers a prioritised merge queue, a full review workspace, and AI review workflows that post inline findings back to the forge, using any OpenAI-compatible model you bring.

What you get¶
-
Merge queue
Every open MR/PR you author, review or watch, in one prioritised inbox. Drag to set your own order.
-
Review workspace
Diffs, threads, drafts, suggestions, approvals and merge checks without leaving Arnold.
-
Story mode
An AI review brief and guided review steps that walk you through a change in a sensible order.
-
AI analysis
Run Improve on a change or a group of changes, with RFC/Jira context, and get inline findings.
-
Review personas
Versioned, publishable review policies (security, performance, architecture…) that run as specialist reviewers.
-
Feedback and insights
Arnold tracks whether authors applied each AI finding, and reports acceptance, token usage and outcomes.
How it fits together¶
flowchart LR
B[Browser] --> UI[arnold-ui<br/>Next.js]
UI -- "/api proxy + session JWT" --> S[arnold-backend<br/>Dropwizard]
S --> A[(Aerospike<br/>state)]
S -.-> E[(Elasticsearch<br/>analytics, optional)]
S <--> F[GitLab / GitHub]
S -- "jobs: local queue,<br/>GitLab CI or GitHub Actions" --> R[arnold-runner<br/>PocketFlow]
R -- "inline comments" --> F
R -- "results + callbacks" --> S
R --> L[LLM<br/>OpenAI-compatible]
Read Architecture for the details.
Where to start¶
| You want to… | Read |
|---|---|
| Try Arnold on your laptop in 5 minutes | Quick start |
| Connect GitHub or GitLab | GitHub App · GitLab |
| Pick the model that writes reviews | LLM and harnesses |
| Run reviews in CI instead of a local worker | Runners |
| Put Arnold in front of a team | Production deployment |
| Look up a setting | Configuration reference |
Arnold is open source under the Apache License 2.0.