Private copy per issue
The team clones your project and only ever writes there. Your folder changes when you Apply.
In development · Beta planned for October 2026
Run an AI team that builds your apps
VividKit is a desktop app for builders who want to ship apps with AI coding agents without living in a terminal. You say what you want; teammates that run on Claude Code work in a private copy of your project, VividKit collects evidence, and the work is done only when you accept it.
Join the waitlist Beta planned · October 2026
Not released yet: there is no download until the beta ships.
Runs on your machine, with your own Claude Code sign-in and AgentKit.
Why VividKit
Coding agents are powerful. Directing them should not need a terminal.
Tools like Claude Code can build real software, but they expect you to type commands, read diffs and trust a "done". VividKit puts a calm desk around them, so you manage outcomes instead of tools.
- Without VividKitPrompts, flags and terminal sessions
You talk, the team works
Describe the change in plain words. Cora picks the right workflow, Ada asks what is unclear, and teammates take turns until there is something to review.
- Without VividKitAn agent saying it is done
Evidence for every criterion
Each Requirement has criteria you can check. For web apps, VividKit screenshots pages at phone and desktop size and records checks, so you accept what you can see.
- Without VividKitAn agent loose in your real folder
Your project stays yours
The team works in a private copy of your project. A restore point comes before every run, and Apply and Publish go through a separate confirmation window.
Who it is for
For builders with an app to build, not a team to manage
You do not need to know Git or write code. You need an idea, or an app that needs work. The beta starts with web apps.
-
Founders
Turn an idea into a working first version, one checked change at a time.
-
Shop owners
Keep your site current: new products, fixes and pages, without waiting on a developer.
-
Freelancers
Hand routine work to the team and spend your time on the parts clients notice.
-
Small teams
One shared AI team for every project, with a clear record of what changed and why.
Take over an app you have
The team scans it read-only, explains how it works, and suggests what to do next.
Start one from an idea
Ada helps you shape the first version and a technical direction before anything is built.
How it works
From one sentence to a published change
You decide at each checkpoint. The team does the work in between, and every step is visible.
- Tell Cora Say what you want in plain words. Cora sorts it and opens an issue in your Inbox.
- Confirm the Requirement Ada asks what is unclear and turns it into criteria you can check.
- The team works Teammates take turns in a relay, in a private copy. Watch or Stop anytime.
- Review evidence Screenshots and check results for each criterion. Accept, or send it back.
- Apply and publish Confirm the exact files in a separate window. Undo and roll back stay possible.
Where it is today
Built and tested in-house. Beta planned for October 2026.
VividKit is not publicly released yet. Here is what works today, what the beta is meant to include, and what comes after. The date is a plan, not a promise.
Works today, in our development builds
- The full loop for web apps: tell Cora, confirm a Requirement, the team works in a private copy, review evidence, Apply, then Publish through Cloudflare Pages.
- Workflows: Bug fix, Quick fix, New feature, New project, Take over a project and Restructure.
- Nine coding-agent CLIs, each of which has run a whole issue in our own tests. Claude Code comes first.
- Test builds run on Linux, on an Apple silicon Mac and on Windows 11.
Planned for the beta, October 2026
- A first public build, 0.1.0-beta.1, offered as a download.
- Linux (.deb and AppImage), macOS on Apple silicon and Windows x64, as unsigned builds.
- An email to everyone on the waitlist when it is ready.
After the beta
- Investigation and Improvement workflows.
- More skill packs, so AgentKit is no longer required.
- Publishing apps that are not websites.
Until the beta ships there is no public download.
Built on
What Claude Code and AgentKit do in VividKit
VividKit is the desk around coding agents, not a model of its own. This is who does what.
-
Claude Code
Does the work · required for the beta
Each teammate, and Cora who sorts your requests, runs as a Claude Code session on your machine, signed in with your own Pro or Max plan.
-
AgentKit
Skill pack · Engineer Kit license required
Stages such as planning, testing and code review use AgentKit skills. AgentKit is a separate product from another team.
-
VividKit
The desktop app
Keeps versioned Requirements, runs the relay in a private copy, takes the screenshots used as evidence, and puts Apply and Publish behind a confirmation window.
-
Other runtimes
Optional
Codex, OpenCode and other agent CLIs can run individual teammates instead of Claude Code.
VividKit does not call the Claude API itself and does not resell model access: runs use your own Claude Code plan. We also use Claude Code to build VividKit. VividKit is an independent project and is not affiliated with, sponsored or endorsed by Anthropic. Claude and Claude Code are trademarks of Anthropic.
The team
Your team, shared by every project
Six teammates with fixed roles, each doing one thing at a time. Rename them, or change the runtime each one runs on.
- CoraCoordinator
Sorts your request, picks a workflow, creates the issue and the first Inbox item. Does not do the work.
- AdaArchitect
Advice, brainstorming, Requirements, plans, project reports and code review.
- DeeDesigner
Layout and style.
- BenBuilder
Sets things up, builds and fixes.
- TessTester
Evidence for every criterion, measurements before and after.
- IvyInvestigator (optional)
Finds causes with evidence. Read only.
Workflows
A fixed path for every kind of work
Cora picks the workflow and you can change it before Start. Each one has checkpoints where nobody moves on without you.
-
Bug fix
Something is broken. You confirm the cause before any fix.
-
Quick fix
A one-file change. If it turns out bigger, the team stops and proposes another path.
-
New feature
Add or change something people see. You can approve the plan first.
-
New project
Start from an idea in an empty folder. You confirm the technical direction.
-
Take over a project
Bring an app you already have: a read-only scan, a report, then suggested work.
-
Restructure
Reorganise the code or move stacks while users see no difference. Tests run before and after.
-
Investigation After the beta
Understand a problem before deciding to fix it. Ends with a report.
-
Improvement After the beta
Make something measurable better, with numbers before and after.
Safety
Built so you stay in charge
Agents move fast. VividKit makes every step visible, reversible and yours to accept.
Evidence, not promises
VividKit screenshots each page at phone and desktop size and checks every criterion.
A separate window for real changes
Apply, Undo, Publish and Roll back run only after you confirm the exact files.
Stop means stop
One press ends the run within three seconds. Nothing already done is lost.
Isolation when Full access is off
Claude Code and Codex then run every shell command behind bubblewrap or Seatbelt, away from secrets. Full access is on by default in the beta.
Requirements have versions
Change your mind mid-way: v2 shows what changed, and v1 evidence turns Stale.
Drawings of how each safeguard works. Not screenshots of the app.
About auto mode and Full access
Both are on by default in the beta. Auto mode lets teammates push, open pull requests, merge and publish if the tools on your machine allow it; turn it off and they only propose. Full access lets every runtime run without operating-system isolation, and the app shows when it is on. Either way VividKit keeps a restore point before each run and offers Roll back.
Runtimes
Works with the coding agents you already use
Claude Code is required for the beta. Each teammate can run on a different runtime; VividKit finds the ones installed on your machine and uses your own sign-in.
- Claude Code
- Codex
- OpenCode
- GitHub Copilot CLI
- Cursor Agent
- Grok
- Oh My Pi
- Pi
- Antigravity
Runtimes other than Claude Code and Codex run only with Full access on.
Before you begin
What you will need
VividKit checks these for you when it starts, and tells you what is missing.
On your machine
- Claude Code, signed in on a Pro or Max plan
- AgentKit with an Engineer Kit license
- A project in a Git folder, or an empty folder for a new app
Systems planned for the beta
- Linux x86-64 Planned (.deb, AppImage)
- macOS, Apple silicon Planned (unsigned)
- Windows 10 and 11, x64 Planned, in testing (unsigned)
- macOS, Intel No build yet
Questions
Good to know
Can I download VividKit now?
Not yet. VividKit is in development and the beta is planned for October 2026. When it ships, everyone on the waitlist gets an email with the download link.
Do I need to know how to code?
No. You describe what you want, answer questions, read the evidence and press Accept, Apply and Publish. Technical details stay one click away for when you want them.
Can the team break my real project?
The team starts in a private copy of your project. VividKit records a restore point before every run, reports any change to the real project or site afterwards, and offers Roll back. Turn auto mode off and only you can apply or publish.
Which AI does it use, and who pays for it?
Your own. VividKit runs the coding agents installed on your machine with your sign-in, so runs use your existing plans. Claude Code on a Pro or Max plan is required for the beta.
Is VividKit made by Anthropic?
No. VividKit is an independent project built on top of Claude Code, which you install and sign in to yourself. It is not affiliated with, sponsored or endorsed by Anthropic.
Is it only for websites?
No, but web apps get the most: page screenshots as evidence and publishing to Cloudflare Pages. For other apps the evidence comes from tests, builds and runs, and publishing is not available yet.
Where do my project files and notes live?
On your machine. Requirements, plans, reports and the changelog are written into your project folder, so you can read them without VividKit.
Bring an app. Meet your team.
The beta is planned for October 2026. Join the waitlist and we will email you when there is a build to download.
You are on the list. We will email you when the beta is ready to download.
About
Who is building VividKit
VividKit is an independent project. Reach us directly; we read every message.
Thieu Nguyen
Founder
Designs and builds the VividKit desktop app, and writes the guides on this site.
GitHub : Thieu NguyenNot affiliated with, sponsored or endorsed by Anthropic, AgentKit or ClaudeKit.
Resources
Free guides and best practices
Separate from the app: guides for AgentKit users and the vibe coding community, in English and Vietnamese. They also cover ClaudeKit, the kit many readers started with.
Some guides link to AgentKit with a referral code.