localreview: Local AI Code Reviews with Claude Code – GDPR-compliant and Cloud-independent

Your code never leaves your machine – and yet you get a structured AI code review. With localreview, you can get a second opinion on your diff in Claude Code based on a locally running model. No cloud provider, no data leak, no dependency on external APIs.

Why localreview?

Claude Code is a powerful tool – but anyone reviewing sensitive codebases faces a dilemma. The built-in AI review leaves the local machine and is sent to cloud APIs. For many companies, this is a deal-breaker, especially with proprietary code, before publication, or in GDPR-regulated environments.

The existing solution from OpenAI (Codex CLI with Codex-Plugin-CC) pointed the way: a plugin that adds a second AI review. But it was tied to Codex – a cloud service.

localreview takes the same approach, but cloud-independent. It sends your diff to a locally running, OpenAI-compatible model – for example, oMLX on your MacBook, or in the future LM Studio.

How it works

The plugin is implemented as a Claude Code plugin and provides five slash commands:

Command	Function
`/localreview:review`	Starts a review of the current diff (with options for scope, branch, model)
`/localreview:status`	Shows running and completed review jobs
`/localreview:result`	Shows the result of a completed review
`/localreview:cancel`	Cancels a running review
`/localreview:setup`	Checks if the local server is reachable and lists available models

The review runs fully asynchronously – large models take several minutes. With --background, you keep working while the review runs in the background.

Local models as the review engine

In the first step, localreview supports oMLX – an LLM inference server specialized for Apple Silicon with continuous batching, SSD caching, and multi-model management. oMLX runs on the MacBook and provides an OpenAI-compatible API at http://127.0.0.1:8000/v1.

The extension to LM Studio is already built into the architecture – since LM Studio also offers an OpenAI-compatible API, only minor adjustments in the configuration are needed.

The crucial difference: data protection

Unlike cloud-based review solutions, with localreview your code never leaves your machine. Communication runs exclusively over localhost:

No data transfer to the cloud
No dependency on external APIs
No storage on third-party servers
Fully GDPR-compliant

The plugin even explicitly warns you if you accidentally configure an external address – your security is built in, not optional.

Installation

# In Claude Code:
/plugin marketplace add mjochum64/localreview
/plugin install localreview@localreview

After that, a simple /localreview:setup checks whether your local server is running.

Configuration

You have three ways to configure the server and model (in this order):

CLI flag: --model <id> on the review command
Environment variables: LOCALREVIEW_BASE_URL, LOCALREVIEW_MODEL
Config file: ~/.config/localreview/config.json

By default, localreview connects to http://127.0.0.1:8000/v1 and selects the first available model.

What makes localreview special

Read-only: The plugin never changes code, never applies suggestions, and runs only on your explicit request
Structured output: Findings are sorted by severity (via JSON schema)
Sensitive files excluded: .env, .pem, SSH keys, and similar files never even make it into the diff
German interface: Commands, error messages, and the review itself are in German
Open source: Apache-2.0 license – viewable, extensible, auditable

Outlook

localreview is the first step in a direction I believe represents the future of code review: local AI as a second pair programmer, independent, data‑protection‑compliant, and always available. Integration of LM Studio, more model backends, and an English interface are the logical next steps.

Anyone who wants to get involved: the repository is open for issues, pull requests, and ideas.

Sources

🌐 Machine-translated from the German original, editorially reviewed. 🤖 Written with AI assistance.