Review layer for AI coding work · Claude Code first

The review layer for Claude Code.

Q: What is Zurge?

Zurge is a review layer for Claude Code. It turns the agent's plans, diffs, and tool output into a reviewable workflow, so you can question the strategy, inspect the implementation, and approve agent-written code before it ships.

Q: Does Zurge replace Claude Code?

No. Claude Code is still the engine that writes the code. Zurge is the review layer around it: review the plan, inspect every diff, send feedback back, and approve before anything ships.

Q: Is my code uploaded anywhere?

No. Zurge is local-first and your code stays on your machine. The optional phone-remote and plan-share features only carry session text or the plan you choose to share, never your repository.

Q: Do I need a new account or API key?

No. Zurge uses your existing Claude Code login. There is no Zurge account and no separate API key to manage.

Q: How much does Zurge cost?

Zurge is free for individual developers.

Q: Which operating systems does Zurge support?

Zurge is available for macOS first. Windows and Linux support is planned.

Zurge turns Claude Code's plans, diffs, and tool output into a reviewable workflow, so you can question the strategy, inspect the implementation, and send feedback back before anything ships.

Join the early access list Watch the review flow

Local-first. No new account. Uses your existing Claude login. macOS first, more platforms later.

claude-code · terminal

⏺ Edit src/checkout/session.ts ⏺ Edit src/lib/api/retry.ts ⏺ Bash npm run build ⎿ 2,140 lines of output (ctrl+r to expand) ⏺ Write src/lib/payments/webhook.ts ⏺ Edit 8 more files... ! changed 12 files, 1 command run ? Plan approved? proceed [y/n]

Terminal output is not a review interface.

zurge · review

◆Planchanges requested

2Change webhook retry policy to exponential backoff

you · commentDo not change retry behavior here. Why this file?

3Add handler in lib/payments/webhook.ts

✎Editsrc/checkout/session.ts

- retry: { max: 5, backoff: "none" }+ retry: { max: 5, backoff: "exponential" } timeout: 30_000,

Zurge turns agent work into something you can review.

The problem

Claude writes fast. Review did not catch up.

Claude Code can generate plans, edit files, run commands, and open PRs faster than a human can comfortably follow in a terminal. That speed is useful, until you realize you are approving work you have not actually reviewed. Zurge exists for the part of AI coding that still needs human judgment: strategy, tradeoffs, risky diffs, test evidence, and the final call.

× Plans get buried in chat.

× Diffs are mixed into noisy tool output.

× Feedback turns into manual copy-paste.

× PR review happens after the agent already went down a path.

× You need a way to steer before the work hardens into something you have to unwind.

The category

Not another agent. The layer above the agent.

Zurge does not replace Claude Code. Claude is still the engine. Zurge is the review layer around it: the place where you inspect what the agent is about to do, what it actually changed, and what needs to happen next.

Claude Code

Runs the work

Plans the approach
Edits files
Runs commands
Shows its reasoning
Implements the change

Zurge

Helps you review the work

Plans and diffs as cards
Comments on lines and hunks
Approvals and change requests
Test and command evidence
Feedback back to the agent

You

Own the decision

Judgment
Taste
Tradeoffs
Risk tolerance
The final call

01 Plan review

Review the strategy before the code exists.

Most agent mistakes start before the first file changes. The plan is too broad, the assumptions are wrong, or the agent picks the wrong path. Zurge makes the plan reviewable.

›Comment on a specific plan line.
›Ask Claude why it chose an approach.
›Request a smaller scope.
›Approve only when the strategy is right.
›Keep the approved plan attached to the implementation.

Catch the wrong approach before it becomes a twelve-file diff.

◆Planrate-limit-loginpending review

1Add a token-bucket limiter in lib/ratelimit.

2Wire it into the auth middleware for every route.

you · ask claudeWhy every route? Scope this to the login endpoint only.

3Return 429 with a Retry-After header.

4Add tests for the limit and the reset window.

02 Diff review

Every edit becomes a reviewable diff.

Instead of scanning terminal logs, Zurge turns Claude's file edits into clean red and green diff cards. Review the implementation while it is still close to the conversation that created it.

›Inline diffs for Edit, Write, and MultiEdit.
›Hunk-level comments.
›Ask Claude about a suspicious change.
›Hide low-signal tool noise.
›Request changes without losing context.

Review agent work like a PR, before the PR.

✎MultiEditsrc/lib/ratelimit.ts

export function allow(key: string) {- return true // TODO+ const b = buckets.get(key) ?? fill()+ if (b.tokens < 1) return false+ b.tokens -= 1; return true }

✎Editsrc/auth/login.ts

+ if (!allow(ip)) return res.status(429)

03 Feedback loop

Turn review comments into agent instructions.

A review tool is only useful if the feedback gets back to the agent cleanly. In Zurge you comment on the plan or diff, then send structured feedback back to Claude in the same worktree. No copy-pasting, no vague "fix this," no lost context.

Comment once. Send it back clearly.

↩Send feedback to Claude

Review feedback

Do not change webhook retry behavior.
Revert the hunk in checkout/session.ts.
Add a test for 429 responses.
Explain why this helper belongs in lib/payments.

↩ Send to Claude

04 Implementation receipt

Know what Claude actually did.

At the end of a task, Zurge creates an implementation receipt: a structured summary that connects the approved plan to the actual diff and the evidence behind it.

Less "looks good to me." More "I know what changed."

▣Implementation receiptneeds changes

taskRate-limit the login endpoint

approved plantoken-bucket, login route only

files changed4 · ratelimit.ts, login.ts, +2

commands runnpm run build ✓

tests not rununit tests for the 429 path

risky changestouched checkout/session.ts retry policy

open questionslimiter scope confirmed for login only?

The workspace

A local workspace for serious Claude Code sessions.

Zurge is a desktop app because agent work is not just a prompt. It is a running session, a branch, a diff, a terminal, a preview, and a set of decisions.

Built around the work

Isolated git worktrees
Multiple Claude Code sessions
Status: working / waiting / needs review / done
Local-first architecture

Stay in control

Raw Claude terminal escape hatch
Live preview for web apps
Uses your existing Claude login
Your code never leaves your machine

Also included

Phone remote to watch and steer
Voice input
Themes
Run a task across several agents

Where it fits

The missing surface between agent output and PR review.

Claude Code is the engine. A parallel agent manager runs more of them at once. Zurge is the layer where a human reviews the work before it ships.

Capability	Raw Claude Code terminal	Parallel agent manager	Zurge
Review the implementation plan	partial	no	yes
Comment on plan lines	no	no	yes
Inspect every diff	in logs	partial	yes
Ask about a diff hunk	no	no	yes
Send feedback back to Claude	retype	retype	structured
Implementation receipt	no	no	yes
Worktree isolation	no	yes	yes
Human approval loop	y/n	partial	yes
Local-first	yes	varies	yes

Claude Code is the engine. Zurge is the review layer.

Direction

Claude Code first. AI agent review layer next.

Zurge starts with Claude Code because that is where the workflow pain is sharpest today. The long-term goal is broader: a review layer for AI coding work, no matter which agent writes it. These are directions, not promises.

→ Support for other coding agents

→ Shared team reviews

→ Approval history

→ Private review links

→ GitHub checks

→ Review policies

Founder note

Built because terminal logs stopped being enough.

"I built Zurge after Claude Code became fast enough to do real work, but the interface for reviewing that work still felt like a terminal log. The more I used agents, the less I wanted more autonomy. I wanted better review, better steering, and more confidence before code reached a PR."

Rajat Kaushik, founder of Zurge

FAQ

Questions developers ask.

What is Zurge?

Zurge is a review layer for Claude Code. It turns the agent's plans, diffs, and tool output into a reviewable workflow, so you can question the strategy, inspect the implementation, and approve agent-written code before it ships.

Does Zurge replace Claude Code?

No. Claude Code is still the engine that writes the code. Zurge is the review layer around it: review the plan, inspect every diff, send feedback back, and approve before anything ships.

Is my code uploaded anywhere?

No. Zurge is local-first and your code stays on your machine. The optional phone-remote and plan-share features only carry session text or the plan you choose to share, never your repository.

Do I need a new account or API key?

No. Zurge uses your existing Claude Code login. There is no Zurge account and no separate API key to manage.

How much does Zurge cost?

Zurge is free for individual developers.

Which operating systems does Zurge support?

Zurge is available for macOS first. Windows and Linux support is planned.

Get started

Let Claude write. Keep the review layer human.

Join the early access list and be first to review your Claude Code tasks in Zurge.

Join the early access list Watch the review flow

Claude Code first. Built to become the review layer for AI coding agents.