Your 1,400 lines are really three things
Diffzu groups large AI-generated diffs into conceptual change-sets with risk levels and review notes, natively in your JetBrains IDE.
Auth refactor
New payments endpoint
- Token TTL dropped 24h -> 15m - confirm the refresh flow.
Formatting noise
The hard part of a big agent diff is not reading the code. It is knowing which parts deserve your attention.
How it works
Run a review
Analyze local changes or your branch against its base. Diffzu reads the diff and, with read-only access, the surrounding code.
Read the map
A handful of conceptual groups with a risk level and a reason. Notes anchored to file and line. Noise collapsed into one group you can skim.
Jump to the diff
Double-click any file to open the native JetBrains diff viewer, with review notes inline.
What ships in 0.5.2
- Local changes or branch vs base
- Claude Code CLI or your own Anthropic API key
- Teach-a-Rule: review preferences as a plain file, per user or per repo
- Explain this: plain-prose explanation of any note
- Review history and markdown export
Your code goes where it already goes
Read-only by design
The review agent gets Read, Grep and Glob. Nothing else.
"a diff is untrusted input"
Your keys, your compute
Runs on your Claude Code subscription or your API key, stored in the IDE password safe. No Diffzu servers, no account, no database.
Telemetry is opt-in
Off by default. Never code, paths, prompts or branch names. PostHog EU.
This site uses cookieless, anonymous analytics. No cookies. Opt out in the footer. Read the privacy policy.
Straight answers
Does my code go to Diffzu's servers?
No. There are none. Analysis runs through your own Claude Code CLI or your own API key - the same path your code already takes when you use Claude Code.
What does it cost?
The plugin is free. You pay only for your own Claude usage.
Which IDEs are supported?
IntelliJ-based IDEs 2025.2 and newer, with the Git plugin enabled.
Does it only work with Claude-generated diffs?
It reviews any diff, no matter which agent produced it. The analysis itself runs on Claude.
Can it change my code?
The review is read-only. Diffzu never edits your files.
I review large agent diffs every day, and the flat wall of changes was the part I dreaded. Diffzu is the tool I built so that the mandatory review of generated code becomes bearable: a map of what changed, with risk where it matters. If it saves your next review an hour, it did its job.
Robert, builds Diffzu