Engineer Setup Guide
Everything your engineer needs to wire up Novawolf node hosting and run the AI in create-mode only.
1. Blockchain Node Hosting Pipeline
The site captures a target website, bundles its HTML plus all linked images, stylesheets, and scripts, and pins the resulting archive to the Novawolf decentralized node network. The pipeline runs through three backend functions:
- —captureSite — scrapes the target domain, downloads every asset, and assembles a self-contained bundle.
- —hostBundle — pins the bundle to the Novawolf Blockchain gateway and returns a content hash plus a live gateway link.
- —verifyDNS — checks whether a custom domain TXT record has propagated and matches the pinned hash.
2. Gateway & Secret Configuration
The hosting pipeline will fail with ERR_NOVAWOLF_NOT_CONFIGURED until the blockchain gateway credentials are connected. Your engineer must set the following in Dashboard → Secrets:
- —Blockchain gateway endpoint URL and API key — required by hostBundle to pin bundles.
- —STRIPE_SECRET_KEY and STRIPE_PUBLISHABLE_KEY — required for the checkout flow on the pricing page.
- —STRIPE_WEBHOOK_SECRET — registered against the stripeWebhook function at https://nodeable.base44.app/functions/stripeWebhook.
3. Domain & DNS Wiring
After a bundle is pinned, the user connects a custom domain. The engineer should confirm the DNS flow end-to-end:
- —The Domain Settings page shows the exact TXT record the user must add at IONOS or GoDaddy.
- —The Verify DNS page polls the live DNS to confirm the record matches the pinned hash.
- —If the gateway returns ERR_NOVAWOLF_BLOCKCHAIN_REJECTED, check that the gateway endpoint is reachable and the API key is valid.
4. AI in Create-Mode Only
The platform uses the InvokeLLM Core integration for AI-assisted problem solving and content creation. The runtime is locked to create-mode only — there is no continuous learning, no training, no backpropagation, and no vector-database injection. The AI generates outputs on demand and never persists or retrains on user data.
- —InvokeLLM accepts a prompt and optional file attachments; it returns a single generated response.
- —Use it to solve problems, draft content, and generate code — never to train, fine-tune, or update a model.
- —Every call is stateless: the model has no memory of previous calls and writes nothing to disk.
5. Inference-Only Guardrails
To keep the AI strictly inference-only, the engineer must enforce these constraints in any AI-backed function:
- —Never store model outputs back into an entity field that feeds a training or embedding pipeline.
- —Never expose raw model weights, fine-tuning endpoints, or training-script access through a backend function.
- —Treat every InvokeLLM response as ephemeral — display it, save it to the user record if needed, but never feed it back as training input.


