-
Notifications
You must be signed in to change notification settings - Fork 0
Application Generation Engine
The current NEXUS AI creation experience is the AI App Builder: a persistent conversation with a live preview, editable files, version checkpoints, GitHub handoff, and deployment. A one-shot Deploy from Prompt flow remains available for simple generation-and-deploy tasks.
| Workflow | Best for |
|---|---|
| AI App Builder | Iterative product work, visual review, direct edits, full-stack apps, and GitHub push |
| Deploy from Prompt | A small service or prototype that can be generated and deployed in one pass |
| Deploy from Git/local source | Existing applications where AI generation is unnecessary |
| MCP builder handoff | Generate files in ChatGPT, Claude, Cursor, or Codex and preview them in NEXUS AI before deployment |
- Open a project.
- Select Build with AI.
- Choose a configured AI provider.
- Describe the application in the chat panel.
- Review the generated files and live preview.
- Continue the conversation until the application is ready.
Builder sessions are persistent. Each assistant turn stores a complete snapshot of the files.
The React preview runs locally in the browser:
- files are bundled with
esbuild-wasm; - the app renders inside a sandboxed iframe;
- the iframe allows scripts but does not receive same-origin access;
- generated code cannot read the NEXUS AI session or local storage;
- previewing does not provision infrastructure.
React single-page apps preview directly. Full-stack Next.js + Prisma apps use a mock-data implementation in preview because the browser sandbox does not connect to a production database. Deployment provisions and connects the real database.
You can change an app four ways:
- Chat — describe the change in plain language.
- Code panel — edit and save a file directly.
- Image attachment — paste or upload a PNG, JPEG, WebP, or GIF as a design reference or bug screenshot. Up to three images can be included in one message.
- Click-to-edit — select an element in the preview, then describe the change. The builder sends a small element description rather than the entire page.
Images are resized in the browser before upload. The selected AI model receives the actual image as a vision input, not only its file name.
- Every assistant response that changes files creates a checkpoint.
- Saving a manual code edit also creates a checkpoint.
- Restoring an earlier checkpoint creates a new checkpoint; it does not erase history.
- Deployment does not end the builder session. Continue iterating and redeploy when ready.
Builder checkpoints and deployment releases are related but distinct. Builder history tracks source snapshots; deployment rollback restores an earlier deployed release.
The builder can create Next.js App Router + Prisma applications. When a full-stack build is deployed, NEXUS AI can provision managed Postgres and inject the database connection.
For built-in authentication, first generate the base app, then ask for login and signup as a follow-up. The builder can add a Prisma user model, password hashing, signed HTTP-only session cookies, and route guards. Review authentication and authorization code before production use.
- Preview build failures can trigger up to two automatic AI repair attempts.
- Deployment status remains visible in the builder header.
- A failed deployment can be sent through Fix with AI.
- If a managed database already exists, the repair redeploy reuses it instead of creating a duplicate.
AI repair is a convenience, not a substitute for tests, security review, or a database backup before risky changes.
- Select Push to GitHub in the builder.
- Connect the NEXUS AI GitHub App if necessary.
- Choose an existing repository or create an allowed organization repository.
- Push the current complete snapshot.
Each push creates a commit containing the current file set, including deletions. Repository creation and push depend on the permissions granted to the GitHub App. Installation tokens cannot create repositories in every personal-account configuration; use an existing repository or an organization installation when required.
An MCP-compatible agent can push an app into a project builder session:
1. nexusai_projects_list
2. The agent generates files in chat or the IDE
3. nexusai_builder_push
4. Open the returned builder preview URL
5. Iterate in the builder
6. nexusai_builder_pull (optional, to return builder edits to the agent)
7. Deploy after review
nexusai_builder_push requires deployments:create; nexusai_builder_pull requires deployments:read.
The platform supports configured Anthropic, OpenAI, Google Gemini, xAI, Cohere, OpenRouter, managed, and compatible custom-provider paths. Exact models and plan availability change over time; use the AI Providers screen as the authoritative list for your organization.
Provider credentials are encrypted. A builder session requires an enabled provider, including sessions created through MCP handoff.
Useful prompts describe:
- the user and primary workflow;
- pages, states, and navigation;
- data entities and relationships;
- authentication and permission rules;
- visual direction, including responsive behavior;
- required integrations;
- success and error states.
Example:
Build a responsive customer-support portal with a dark blue visual system.
Users can sign up, create tickets, attach screenshots, and view ticket history.
Admins can assign tickets and change status. Use Next.js + Prisma and include
empty, loading, validation, and permission-denied states. Start with the base
app; we will add authentication in the next turn.
Prefer focused follow-ups over asking one model response to implement every feature at once.
- Review all generated dependencies and code.
- Add automated tests for critical behavior.
- Verify authentication, authorization, and tenant boundaries.
- Keep credentials in the Secrets Vault.
- Confirm database migrations and backup strategy.
- Test health checks, startup time, and shutdown behavior.
- Review the public URL, custom domain, and TLS state.
- Check logs after the first production deployment.