DiffGuardian：用语音和可视化讲解任意 PR 或代码库

Speaking

…then authMiddleware asks PermissionService whether that role can perform the action, and answers with a 403 if it can't.

HoldSpaceto askReplay ↺

Talk to your PR

Get a live walkthrough of any PR, out loud, right now.

Hold Space and ask what a PR does. DiffGuardian walks you through it as it goes: the features it ships, what each one changes, what it touches. You hear it while the code it's explaining scrolls into view and lights up. Ask a follow-up whenever you like.

• Instant. No setup per PR. Ask and it starts talking.
• Live. The code it's explaining is highlighted and scrolled into view as it talks.
• Yours to steer. Ask for the part you care about first, or skip straight to the one file that matters to you.

Pro. Runs on your own OpenAI or ElevenLabs key, or your system's built-in voice.

Speaking

…this hunk replaces the hard-coded admin check with a list of actions for each role, so editors can publish but only admins can delete.

HoldSpaceto askReplay ↺

Talk to your codebase

Ask the whole repo a question. Hear the walkthrough. See it on a live graph.

DiffGuardian, available on macOS, Windows and Linux, indexes your repo into a code knowledge graph. Hold Space and ask how auth works or where a webhook is verified. It answers out loud while a live visual of the flow builds on screen, straight from your call graph.

Speaking

…then authMiddleware asks PermissionService whether that role can perform the action, and answers with a 403 if it can't.

HoldSpaceto askReplay ↺

History · 8

You saidHow does the app decide who can reach the admin routes?

Spoken · 0:22

Every request passes through authMiddleware. It verifies the token, loads the user's role, then asks PermissionService whether that role is allowed to do the action. Admin routes require the admin role. Anyone else gets a 403, and the denial is logged.

▤ Role-based access · on canvas

You typedWhere are the roles defined?

1. 01Ask out loudHold Space and ask. No typing, no prompt writing.
2. 02Hear it explainedPlain English, with the real files named as it goes.
3. 03See the graphThe visual aid builds as it talks.

Prefer typing? The chat is still there.

Trace It

Prefer to read? Trace It gives you the walkthrough at your pace.

Hit Trace It! and a 5,000-line PR becomes a guided tour, one feature at a time, each explained in plain English. Read it as slowly as you like, go back, and approve having seen the whole thing, not skimmed a diff.

• Walkthrough, written — the PR broken into the features it ships, each explained in plain English
• Blast radius — a dependency graph built from the diff alone (heuristic, not AI): every file and module a change actually touches, before it merges

Want it spoken instead? Talk to your PR.

Built for parallel agents

Ten agents, forty PRs. You still know what's shipping.

When agents work in parallel, the bottleneck is you. For every PR, DiffGuardian finds the quickest way for you to understand it, so you spend minutes on each one instead of reading every line or rubberstamping it.

• Read itA written walkthrough, feature by featureTrace It breaks the PR into what it ships and explains each part at your own pace.See how
• Hear itThe same walkthrough, spokenAsk out loud and get it told back, with the code it's explaining highlighted as it goes.See how
• See itA graph of what a change touchesEvery file and module a change reaches, drawn from the diff alone, before it merges.See how

Who it's for

For everyone who needs to know how their products are actually built.

EngineersUnderstand a PR well enough to approve it, without reading every line.

PMsAsk what shipped and get a plain-English answer you can trust, without reading code or waiting for someone to explain it.

Anyone building softwareFounders, designers, QA, analysts, anyone working on an application: ask what the code does and get a plain-English answer, spoken.

Working with a team?

Onboard in minutesA new developer sets up in a few minutes, then talks to the codebase and hears how auth or billing works, with a live graph.

Everyone understands the PRThe author, reviewer and PM each get the walkthrough that suits them: read it, hear it or see it.

PMs ask, engineers keep buildingA PM asks out loud how a feature works and gets a plain-English answer.

Hand-off without the authorInherit an agent-written PR and get the walkthrough instead of waiting on someone's memory.

Your voice, your keys

Your models. Your keys.
Your voice.

Your code never touches our servers.

Bring your own key for Anthropic, OpenAI, Gemini or OpenRouter, or run a local model. Pick a model for each job. Voice runs on your own OpenAI or ElevenLabs key, or your system's built-in voice, and none of it reaches our servers.

• One OpenRouter key reaches ~400 models, with no new account per vendor
• You pay the provider directly, with no markup
• Ollama, LM Studio, llama.cpp and vLLM work out of the box
• Knowledge bases stay on your machine, and credentials sit in your OS keychain

Routing this review through…

Your machinellama-3.3-70b · local

Anthropicnot contacted

OpenRouter~400 models · not contacted

0bytes left your machine

0tokens in cloud cache

100%keys held in your OS keychain

How DiffGuardian compares

Most tools take you out of the code. DiffGuardian keeps you in it, and makes it quick.

The goal isn't less code in your head. It's the same understanding in a fraction of the time. See the full comparison.

Pricing

Free to start. Upgrade to talk to your code.

Every plan works with local models. Pro adds voice and cloud providers on your own keys. Enterprise, with single sign-on, is coming soon.

Downloads

USD$8/mo

For engineers who want to move fast.

Get Pro

• Everything in Basic
• Voice: talk to your PRs and your codebase
• Unlimited repositories and AI reviews
• Managed AI Providers — BYOK, incl. OpenRouter
• Trace It, Ask, Explain and Polish
• Analyze Project, knowledge bases and RAG
• MCP Servers integration
• And many more…

For teams

USD$8/seat/mo

For teams moving fast, working together.

Register my team

• Everything in Pro, for everyone
• Centralised seat billing for the whole team
• Seat management and invoicing
• Minimum of 3 seats

Set up by an organisation admin, who buys the seats for the team. Everyone else just downloads the app — individual sign-in happens inside DiffGuardian during onboarding, never here.

Compare plans

Every feature, side by side.

Questions

Your questions, answered.

What does DiffGuardian actually do?

DiffGuardian helps engineers, PMs and anyone building software

understand code as fast as agents write it. It gives you a walkthrough of a pull request one feature at a time, lets you talk to a PR or your whole codebase and hear the answer, and leaves the decision to approve with you.

Can I really talk to a PR?

Yes. On the PR page, hold Space, ask, release, and hear the answer while the lines it discusses are highlighted. Analyze Project has the same push-to-talk in a full-screen overlay, with a graph that builds as the answer is spoken. Voice is a Pro feature.

What's the difference between Trace It and voice?

Trace It turns a PR into a written, feature-by-feature walkthrough that you read at your own pace. Voice (Pro) gives you the walkthrough out loud, live, as soon as you ask, and lets you steer it by asking questions. They are separate features.

Can it keep up with many agents working in parallel?

It is built for that queue: for each PR, DiffGuardian finds the quickest way to help you understand it, whether that is a written walkthrough, a spoken one with the code highlighted, or a graph of what it touches. It speeds up understanding, you approve when you're ready, and it never merges anything.

I'm a PM, not an engineer. Is it for me?

Yes. Ask what a PR or your codebase does and get a plain-English answer out loud, with the code highlighted or a graph of the codebase alongside. You don't need to read code to follow it.

Is it only for solo developers?

No. A new developer can set up in a few minutes and talk to the codebase to learn how it works. The author, reviewer and PM can each get the walkthrough that suits them, and whoever inherits an agent-written PR can get it without waiting on the author.

Where does my voice go?

Straight from your machine to the provider you chose, using your own OpenAI or ElevenLabs key, or it stays with your operating system's built-in voice. Nothing reaches DiffGuardian's servers.

Is my code private?

Your code never reaches a DiffGuardian server. Reviews run on your own machine, and cloud reviews go directly from your machine to the provider whose key you supplied — under your own agreement with them. Because local models are supported on every plan, you can use the AI features without a single line of code leaving your laptop. See the

Security page for the full data-flow breakdown.

Which platforms is DiffGuardian available on?

DiffGuardian is available for macOS, Windows and Linux. The Windows build is an x64 installer that isn't code-signed yet, so Windows may show a SmartScreen warning before you can run it. On Linux, install the .deb or

.rpm package for your distribution — they declare the dependency DiffGuardian's secure key storage needs, so it works out of the box. Use the AppImage only if neither package fits your distribution; it still auto-updates like every other build.

How does billing work, and can I cancel?

Pro is USD $

8

a month, or $

80

a year — two months free. Team is the same price per seat: $

8

a seat a month or $

80

a seat a year, with a minimum of

3

seats. Cancel any time from the billing portal — you keep access until the end of the period you already paid for.

How is this different from the AI tools I already have?

DiffGuardian doesn't judge the code for you. It makes sure the person who approves understands it. See how DiffGuardian compares.