DifVi is a Mac workspace for AI-written code. Explore changes as a map, inspect the code and running app, and send targeted feedback to your coding agent.
Mac appFree during the closed betaDetails by email right awaySee the workflow
Works with Claude CodeCodexCursor
Agents change hundreds of files in minutes. You still have to answer for every one of them.
The questions are simple. Finding the answers in a 200-file diff isn't.
DifVi maps the change and walks you through it in the order the data flows. Then it checks what could break, and sends the fix back to your agent.
Share it internally so your team quickly knows exactly what's shipping. Ask why about any file, or highlight multiple lines and ask What or Why.
w a tour of the change, in flow order
When your agent reviews its own work, it only reads the diff. DifVi also reads the code around it: the code that calls what changed, the code that writes the same tables, and your team's rules. So it catches what breaks outside the diff: security holes, lost data, wrong charges and broken callers. It shows you where to look, and you make the call.
Before you merge
redeem() checks redeemedAt, then charges the order before marking the card used. Two checkouts at the same moment both pass the check.
Worth checking
refundOrder() reverses the payment but leaves the gift card balance spent.
findByCode() looks codes up as typed, while new codes are stored in capitals.
Comment on a line, or message the agent about the whole change. Your note goes to the agent that wrote the code, and you review its revision in the same place.
Steps: you comment on a line; the comment is delivered to the agent session that wrote it, via hooks, for Claude Code, Codex or Cursor; the agent fixes the code; the issue clears and the map refreshes.
Open the running app inside DifVi. Select an element or an area, write a note, and your agent gets it with all the context available: the page, the element and the code behind it.
localhost:3000/checkoutdiv.total-row “Total $50.00”src/components/OrderSummary.tsx:42Keep track of every agent you have running: see which ones are working, which are waiting on you, and jump between them from the keyboard.
All your worktrees and agents in one list. Menu bar, notifications and a dock badge when one is waiting.
] jumps to the next unviewed file, in flow order. A fix takes four keys.
Your agent still writes the code, on your own account. DifVi shows you what it did and carries your feedback to it.
ServersNext.js, Remix, Nuxt, SvelteKit, Express, Fastify, Hono, Koa, NestJS, tRPC
InterfacesReact, Vue, Svelte
DataPrisma, Drizzle, TypeORM, Sequelize, Knex, Kysely, MongoDB, Supabase, Postgres, MySQL, SQLite
JobsBullMQ, Inngest, Trigger.dev, pg-boss
ServersDjango, FastAPI, Flask
DataDjango ORM, SQLAlchemy
JobsCelery, RQ, Dramatiq
ServersGin, Echo, Chi, Fiber, Gorilla
DataGORM, sqlc, database/sql
JobsAsynq, River
DifVi AI is built in: sign in and the summaries, walkthroughs, and What/Why just work. Local model support comes with DifVi Enterprise.
Evaluate DifVi with your own engineers, codebase and requirements. It starts with a short conversation with us.
Yes. Every file opens as a normal diff. The map shows how the files connect and the order to read them in, so you know where to look first.
Yes. DifVi doesn't write code. Your agent (Claude Code, Codex or Cursor) writes it, on your own account. DifVi shows you what it changed and sends your feedback back to it.
Claude Code, Codex and Cursor, on macOS (Apple silicon and Intel). The flow map covers TypeScript and JavaScript (Next.js, Remix, Nuxt, SvelteKit, Express, Fastify, Hono, Koa, NestJS, tRPC, with React, Vue or Svelte), Python (Django, FastAPI, Flask) and Go (Gin, Echo, Chi, Fiber, Gorilla), plus SQL and GraphQL. Other languages aren't supported yet. Any git repository, branch or pull request opens as a diff.
The map, rules and issues never use AI and are worked out on your Mac. AI features send the code they work on to DifVi AI, whose model providers are set to zero data retention and never train on your code. For local models, ask about DifVi Enterprise.
Yes, with DifVi Enterprise. The AI features can run on a local model you host instead of DifVi AI, so the code they read never goes to us. The map, rules and issues never use AI on any plan.
You make the call; DifVi makes the review easier. It maps the change, explains it with AI when you ask, and its risk checks point out what could break before you merge. Your feedback then goes back to the agent that wrote the code.
Yes. Click + Review in DifVi to open any branch or pull request, compared against its real base.
DifVi is in closed beta. Leave your email to request access. We read every request and reply by email. There's no pricing yet.
Available on
Mac appFree during the closed betaDetails by email right awaySee the workflow
Your access request has been received.
We've emailed you the details. Check your inbox.
Thanks, that's saved. We read every answer.
