How Vivd Works
Vivd is not just a prompt-to-template generator. It runs an AI coding agent inside a real website workspace, then keeps preview, publishing, and self-hosting much simpler than a typical custom-code setup.
If you want a rough mental model, Vivd is closest to:
- a WordPress-like focus on getting sites live without a lot of ops work
- a Lovable-style AI-native editing experience
- but backed by actual project files and real development tooling
Core idea
Each project has a real workspace, not a hidden internal representation. The AI agent works against that workspace, which is why Vivd can be much more flexible than a boxed-in template builder.
In practice, that means Vivd can work at the same level a developer would:
- edit source files
- update layouts, components, content, and styles
- use framework tooling
- install dependencies when needed
- run project-specific workflows inside the isolated runtime
Studio is more than chat
Vivd Studio is not only an agent conversation view. It combines several ways of working on the same project workspace:
- agent-driven edits through OpenCode
- direct text selection and in-preview editing
- an asset and file explorer for uploads, organization, and file-level changes
- a live preview canvas tied to the current workspace state
That matters because Vivd is not trying to force everything through prompts alone. You can mix higher-level AI changes with direct adjustments in the same environment.
Isolated Studio machines
When you open Studio, Vivd does not drop every user into one shared editor process. It ensures an isolated Studio runtime for that project workspace context.
That isolation is one of the main reasons Vivd can be both powerful and safer:
- project work is separated from other projects
- runtime state can be stopped and restarted independently
- edits happen inside a dedicated environment instead of a global shared shell
The underlying runtime provider depends on deployment, but the product model stays the same: isolated Studio environments around real project workspaces. On single-host self-host setups, that usually means Docker-isolated Studio machines. On larger hosted or managed setups, it can mean a provider such as Fly Machines.
What runs inside that Studio runtime
The isolated Studio environment is effectively the working machine for the project. It is where Vivd runs:
- the real project workspace
- the OpenCode server used by the agent
- the project dev server when the project needs one
- the preview-serving routes that feed the Studio UI
So when you look at the page inside Studio, that preview is not a fake mockup. It is coming from the isolated runtime that is currently holding the workspace and serving the project back into the Studio shell.
OpenCode inside the workspace
Vivd runs OpenCode inside the Studio runtime and wraps it with Vivd’s own Studio UX and product-specific tools.
That gives you two layers at once:
- general coding power inside the real project workspace
- Vivd-aware tools for things like plugin info, publish-checklist context, and image flows
This is why the agent can feel much less constrained than a visual-only builder. It is operating inside the actual project environment, not inside a narrowed block editor.
Astro-first Studio workflow
The newer Studio-led “start from scratch” flow works from an Astro + Tailwind starter workspace, and Vivd keeps that project as Astro instead of replacing it with a private internal format.
That means the AI is shaping a real project structure with real source files.
Vivd also supports generic/static projects, but the current Studio-first creation flow is increasingly centered on Astro workspaces because they offer a strong balance of:
- modern component-based structure
- static-site performance
- flexible content/layout composition
- straightforward preview and publish behavior
Bucket-backed persistence
Vivd treats object storage as the durable source of truth for project artifacts.
At a high level, Vivd stores:
- source artifacts
- preview artifacts
- published artifacts
- related OpenCode session state needed for the Studio experience
When a Studio machine starts, it can hydrate the workspace from object storage. When work finishes or sync is triggered, it can push the latest workspace state back out.
That design makes restart and reuse much more practical because the project does not only live on one machine’s local disk.
Out-of-the-box self-hosting
Vivd is opinionated about making self-hosting practical. The solo self-host bundle can
bring the whole stack up for you, including Caddy, frontend, backend, scraper, Postgres,
a local S3-compatible bucket via MinIO, and Docker-based Studio machines.
So if you want everything on your own server, Vivd can provide a full working default without requiring separate managed storage on day one.
Local bucket or external bucket
You have two main storage directions:
Bundled local S3-compatible storage
This is the easiest self-host path. It keeps storage inside the stack with the bundled MinIO/local S3-compatible setup.
Good fit for:
- single-host installs
- smaller self-host deployments
- people who want the fewest external dependencies
External S3-compatible storage
If you want a more scalable or already-standardized storage setup, Vivd can use an external S3-compatible bucket instead.
That includes options like:
- generic S3-compatible storage
- Cloudflare R2
- other external bucket providers exposed through S3-compatible APIs
Good fit for:
- larger self-host installs
- teams that already have object-storage infrastructure
- setups that want storage independent from the main app host
Docker or Fly
Vivd supports more than one runtime path for Studio machines.
Docker provider
The Docker provider is the practical single-host/self-host path. It keeps Studio runtimes on the same machine while still isolating them from the main app process.
Fly provider
The Fly provider is the scalable machine path. It is the production-oriented model when you want Studio runtimes to scale more like dedicated app machines instead of sharing one host.
The important point is that Vivd keeps a similar product model across both:
- isolated Studio runtime
- real project workspace
- object-storage-backed persistence
- managed preview/publish flow on top
Why this is powerful
Vivd is flexible because it is not forcing the AI to stay inside a tiny set of allowed theme settings or page blocks.
The agent is working in a real project with real tooling, which means it can handle work like:
- structural redesigns
- component and style refactors
- Astro component edits
- content and asset updates
- plugin-related implementation work
- publish-readiness cleanup
That is the reason Vivd can feel much more open-ended than traditional site builders.
Why this is still easier than rolling your own
Even with that flexibility, Vivd is still trying to remove the annoying platform work.
It brings together:
- isolated runtimes
- project hydration/sync
- preview and publish handling
- domain and self-hosting workflows
- install-time defaults that work out of the box
So the goal is not “developer power at the cost of simplicity.” The goal is developer-grade flexibility with a much easier path to getting the site live.
Related pages
- Read Getting Started for the quickest first-project path.
- Read Self-Hosting for the solo install flow.
- Read Instance Settings for install-wide runtime defaults.
Read Domains & Publish Targets for domain and launch rules.