AI Website Building and Automatic Content Upload

In this guide: configure your AI tool and model, create an empty project, send one startup prompt, then follow the AI's questions to provide materials, approve samples, and complete the website.

AI does not create accurate website content from nothing. Treat it like an assistant that can read files and operate the admin panel: give it source materials, a target site, and clear instructions, then review the frontend result.

New users should start with the skill route in Step 6. If ZCode, your model, and access are ready, go directly to Step 6: Upload After Installing the Skill. Step 5 is an optional docs-only route; you do not need to do both.

Workflow

1Choose a toolZCode first

2Configure modelBigModel or third-party model

3Grant accessFiles, browser, internet

4Prepare materialsFolders, spreadsheets, images, or old site URL

6Install the SkillRecommended: empty project, one startup prompt

7Confirm and reviewSample first, then complete upload

1. Start With ZCode, Use Codex as Optional

For most beginners, start with ZCode. Its installation, sign-in, and model setup are more localized for domestic model workflows, and the steps below are easier to follow.

Codex can also work, but this page does not focus on Codex installation. Any AI tool that can read files, access the internet, open a browser, and operate AllinCMS site admin can follow the workflow below.

Tool	Priority	Best for	Notes
ZCode	Recommended	Users who want to use BigModel / GLM or third-party models in a domestic model workflow	Setup is more localized; beginners should follow the ZCode steps below first
Codex	Optional	Users who already have Codex installed and can access GitHub and AllinCMS admin	This page does not cover Codex installation; use the upload SOP below if Codex already works

If you are a beginner, follow the ZCode route by default.
If Codex already works reliably on your computer, continue from the access section.

2. Install ZCode and Configure the Model

The ZCode route has three parts: install ZCode and sign in with BigModel, enable the recommended settings, and optionally connect a third-party model such as DeepSeek. The three reference guides are reorganized below as one beginner-friendly setup flow.

References:

• ZCode installation and BigModel sign-in
• ZCode recommended settings
• Connect a third-party model in ZCode: DeepSeek example

1. Download and Install ZCode

Open the ZCode Chinese website and download the version for your computer. If you are not sure which build to choose, use the recommended download option on the official page.

System	Installation
Windows	Download the Windows installer and follow the prompts
macOS	Open the .dmg file and drag ZCode into Applications
Linux	Choose AppImage, DEB, or RPM

On first launch, click Start using ZCode. If you do not need to migrate old data, skip migration and go straight to the login options.

2. Connect BigModel

On the welcome page, choose Connect BigModel to continue. A browser authorization page will open. Sign in with your BigModel account and approve the authorization.

After authorization, return to ZCode and open Settings -> Model Settings. The setup is complete when:

• BigModel shows Enabled.
• The page shows your current plan or quota.
• Models such as GLM-5.3 and GLM-5.3-Flash appear in the model list.

3. Enable the Recommended Settings

For beginners, start with a stable and simple setup. At this stage, the goal is simple: the model works, local files are readable, the browser can be controlled, and long-running project context can be reused.

Setting	Recommendation
Plan	Start with the built-in BigModel plan; check current pricing before upgrading
Daily model	Choose BigModel -> GLM-5.3-Flash
Access	Enable Full access
Computer control	Enable computer control and the input-box computer-control button
Browser control	Enable built-in browser control; import Chrome login state only when needed
Memory and index	Enable workspace memory, new-folder indexing, and repository indexing

Plan: for testing, start with the built-in plan. Before paying, check the current pricing and benefits in ZCode rather than relying on an old screenshot.

Model and access: when creating a new task, choose BigModel -> GLM-5.3-Flash below the input box, and set access to Full access. Website content upload involves reading local files, opening pages, running tools, and checking the frontend, so limited access often gets in the way.

Start with one stable model

For reading materials, uploading products and posts, and checking screenshots, a fast multimodal model such as GLM-5.3-Flash is usually enough. Switch to stronger models only for complex long-running tasks.

Computer control: open Settings -> Computer Control, enable computer control, and show the computer-control button in the input box. If your system asks for accessibility, screen recording, or similar permissions, grant them as prompted.

Browser control: open Settings -> Browser Control and enable built-in browser control. Import Chrome login state only if you want the built-in browser to reuse websites you have already logged into.

Memory and indexing: open Settings -> Memory and enable workspace memory. Then open Settings -> Indexes and enable new-folder indexing and repository indexing. This helps AI understand your project folder and resume future work without starting from zero.

After setup, check these items before starting a website upload task:

• The model is GLM-5.3-Flash or another model you have confirmed works.
• Access is set to Full access.
• The input box shows the computer-control button.
• Built-in browser control is enabled.
• Workspace memory and indexing are enabled.

4. Optional: Connect a Third-Party Model

To use DeepSeek or another provider compatible with OpenAI / Anthropic-style APIs, open Settings -> Model Settings, choose an existing provider, or click Add provider.

DeepSeek: Top Up and Get an API Key

If you want to use DeepSeek models, first sign in to the DeepSeek open platform. Depending on your account status, the platform may ask you to complete real-name verification, top up your balance, or create an API key. Follow the current page prompts.

For beginners, check it in this order:

1. Sign in to the DeepSeek open platform.
2. If the page asks for real-name verification, click the verification entry and complete it first.
3. Open Top up, choose a small test amount such as ¥10, ¥20, or ¥100, then pay with Alipay or WeChat Pay.
4. After the top-up succeeds, open API keys.
5. Create a new API Key, copy it once, and save it only in your own password manager or ZCode settings.
6. Return to Settings -> Model Settings in ZCode and paste the key into the DeepSeek provider configuration.

Official DeepSeek links:

• DeepSeek API Keys
• DeepSeek API Docs
• DeepSeek API Platform

Choose the model ID this way:

Model ID	Best for
deepseek-v4-flash	Daily copywriting, source cleanup, batch text work, lower cost and faster responses
deepseek-v4-pro	More complex planning, long-form organization, code, and multi-step tasks
deepseek-v4-flash-vision-exp	Reading screenshots, page images, and table images

If the AI needs to read screenshots, use deepseek-v4-flash-vision-exp and enable image input for that model in ZCode.

You usually need:

Setting	What to enter
Provider	For example DeepSeek, or your model provider name
Base URL	The provider endpoint, for example https://api.deepseek.com
API format	The format supported by the provider, such as Responses
API Key	Your provider API key; do not share it publicly
Model ID	The exact model ID from the provider

Using DeepSeek as an example, the provider page normally includes Base URL, API Key, API format, and a model list. Provider names and exact fields may vary, so always follow the provider's current documentation.

If the model should read screenshots, enable image input for that model. Otherwise, the AI may not be able to understand uploaded screenshots.

Keep API keys private

A third-party model API Key can spend your account balance. Enter it only in ZCode settings. Blur or remove the key before sending screenshots, recordings, or support messages.

3. Grant Access

Can you access the internet now? Please open GitHub and https://workspace.laicms.com to check.

Whether you use ZCode, Codex, or another AI tool, backend upload requires enough access: local files, internet, and browser operation.

If you use Codex, choose Full access:

Backend upload needs enough access

For small copywriting tasks, limited access may be enough. For reading folders, logging into the admin panel, uploading images, publishing posts, and checking the frontend, the AI needs file, internet, and browser access. If you are not comfortable granting that, use AI only for content preparation and do the admin operations yourself.

4. Prepare Real Content

Keep website materials in a dedicated folder, such as Website Files on the desktop. Website Files is only an example folder name; in real use, the AI should work from the folder you drag into the chat. Do not mix unrelated IDs, contracts, private bills, or customer-sensitive files into that folder.

AI can upload content only when you provide content.

You can provide:

Source	How to give it to AI	Useful for
Local folder	Drag the folder into the chat or give the path	Products, posts, images, logo, datasheets
Excel / CSV	Drag the file into the chat	Product names, specifications, categories
PDF / Word	Drag the file into the chat	Product introductions, company profile, article drafts
Old website URL	Paste the URL	Copy old-site content into a new site
Image folder	Drag the folder into the chat	Product images, post images, gallery images

Prepare your materials now and provide them when the AI asks after installation in Step 6. Keep source materials separate from the empty project folder; do not put client materials in the public tool repository.

If information is missing, have the AI list it and provide it before continuing. Do not let it guess specifications, certifications, customer cases, or contact details.

Only migrate your own old website or authorized content. Replace unauthorized images and text with your own or properly licensed materials; AI rewriting does not grant permission to use them.

5. Upload by Letting AI Read the Docs First

Optional: let AI read the docs without installing a skill

Send the AllinCMS docs to the AI. Ask it to read the relevant guides and the global AI SOP at the bottom of the workflow page, then prepare an inventory and operation plan.

This route helps you understand the workflow or organize materials. Reading documentation does not install tools or prove the AI can operate the admin panel; uploading still requires working tools, a target site, and your authorization.

For the guided website-building workflow, use Step 6 below. Both routes require material confirmation and sample review before completing the full inventory.

6. Upload After Installing the Skill

Recommended starting point

Think of a skill as a website-building guide and toolkit for AI. Send the full startup prompt below, wait for installation, then answer the AI's questions and confirm each stage. You do not need to write your own upload commands.

1. Create an Empty Folder Project

1. Create an empty folder on your computer, for example AI Website Project.
2. Create or open a project in ZCode and select this folder. In another AI tool, use its equivalent local project option.
3. Start a new conversation in that project, with your model and access already configured.

This folder holds the tools and project records. For later changes to the same website, reopen this project and continue the original conversation instead of creating a new project each time.

2. Send the One-Click Startup Prompt and Wait for Installation

Send this entire code block to the AI in one message. Let the AI handle clone, installation, and self-tests; you only need to replace the blank task field with your real request.

[B2B Export Workbench - One-Click Start] Copy this entire block to your AI.
Repository: https://github.com/tony-apan/b2b-export-ai-workbench-source.git
Step 1: Clone the repository, install the website toolkit, and run its self-tests:
python3 sub-libraries/website-content-ops/SKILL-INSTALL/install.py
(On Windows, use install.cmd.) If Node or Python is missing, give me commands I can copy; do not skip them.
Step 2: My task is: _____ (examples: build a website / write a B2B article / export growth advice).
Rules: For website tasks, read sub-libraries/website-content-ops/; for methods, start with wiki/index.md.
Before deleting, overwriting, or publishing, list the operations and wait for my item-by-item approval.
Keep credentials only in environment variables. Ask when something is unclear; do not guess.

Repository: B2B Export Workbench.

If Node, Python, or another required tool is missing, follow the copyable commands the AI gives you, then continue setup and self-tests. Continue only after installation completes and self-tests pass. Downloading the repository alone is not enough.

Screenshot example: create a ZCode task and paste the one-click startup prompt

Create a new task or open the empty project first, then send the full startup prompt above to the AI. The screenshot only shows the operation position; use your own project name, model, and task details.

3. After Installation, Describe the Task in Plain Language

After installation or repository reading, do not study every file yourself. Tell the AI what you want to do, and it will route the task to the right method:

• Build or update a website: Give it customer materials (PDF, Word, spreadsheets, website links, and images are all acceptable) plus the AllinCMS account. Let the AI ask follow-up questions and continue the workflow.
• Write B2B English articles: Give it product and company materials. Ask for an outline, fact checklist, and missing information before drafting the article.
• Export growth consulting: For cold email, LinkedIn, SEO, trade shows, and similar tasks, explain the industry, target market, current situation, and problem to solve.

For website building or updates, answer the AI's later questions with the admin address, account, password, customer materials, and special requirements. Enter credentials locally as instructed, rather than putting them in public screenshots or project files. After the AI organizes the materials, have it return an inventory for your confirmation before moving on.

Conversation example: provide account information and materials

Read what the AI needs and provide your own information item by item.

4. Review the Complete Inventory First

Let the AI finish reading the materials and produce an inventory. Check three things:

1. Nothing is missing: company information, all products, posts, categories, tags, images, contact details, counts, and sources.
2. Gaps are explicit: missing specifications, images, brand copy, and contact details must be listed separately, not invented or marked complete.
3. Planned changes are clear: the target site, first product and post samples, pages to handle later, and any deletions, overwrites, or publishing in this round.

Reply to numbered items individually. State which you approve and what needs changing. Before approving deletion, overwriting, or publishing, check the targets and counts; a general agreement must not become permission for every future operation.

Conversation example: confirm the website and migration scope

Notice how the AI lists questions and the user answers by number. Login choices, article counts, and replies belong to that task; do not copy them into your project.

5. Review Product and Post Samples Before Continuing

After confirming the inventory, start with one product sample and one post sample, using the content types actually supplied. Do not invent materials just to create a sample. Products and posts require publishing for preview: approve the specific sample publication first, then have the AI return backend records and frontend view links.

Open the links and check copy, specifications, categories, tags, images, and layout. Send screenshots and clear corrections when needed. Once the samples meet your requirements, confirm that the remaining content should follow that format.

Conversation example: approve the sample plan and start the first batch

Confirm sample content, images, and publishing scope before execution. You do not need to repeat technical terms in the screenshots; focus on the content that will change.

6. Follow the Prompts Until the Whole Website Is Complete

Once samples pass review, have the AI list the remaining content and this round's changes. Continue within the approved scope, including the homepage, menus, company information, and contact details. Keep reviewing results, answering questions, and confirming the next round.

Final results must reconcile with the original inventory. If it contains 50 products, report the result for all 50, not just a few uploads. The same applies to categories, tags, posts, and images. List unfinished items with reasons and missing materials.

Every image displayed on the website must be uploaded to the AllinCMS admin media library, then linked to its product, post, or page. Pasting an old-site or external image URL is not a substitute.

After each round, have the AI save completed items, view links, unfinished work, the approved sample format, and the next step in private local project records. Reopen the original project next time. Update records when the process changes to prevent duplicate uploads.

Conversation example: review the result and complete the remaining content

Review the current page, then confirm brand copy and subsequent posts. Continue with products, the homepage, and company pages. Example replies approve only the items listed at that time; use your own inventory.

What counts as complete?

The website is complete when all approved content is finished and verified, with working website links, counts, and check results returned. If anything is missing or has failed, report partial progress rather than claiming completion. Tool installation and sample uploads are intermediate milestones. Continue with the website review in Step 7.

Review the links, resolve missing items, then check the final delivery inventory. A 200 status in the AI's report normally means the page request succeeded; it does not prove that specifications, images, or page content are complete. Have the AI fill gaps from the original materials and check again.

Conversation example: open review links and check for missing content

In this conversation, the AI returns links to the homepage, products, posts, company information, and contact page while noting that some product specifications still need completing. Open the links, review the result, and confirm what to fix. Missing items mean this is only a completed stage, not a completed website.

Conversation example: review the final report after filling the gaps

After completing the missing content, ask the AI for a final report covering brand and template updates, products and posts, specification tables, pages, backend media, and frontend checks. Compare it with the original inventory, including categories, tags, and images.

The product, post, and page counts in this screenshot belong to this example project; they are not a fixed website-building requirement. Review every item in your own approved inventory. The final report must also distinguish completed work from remaining tasks.

7. Confirm Before Upload, Review After Upload

After Step 6, open the website links returned by the AI and review the following:

Check	What to verify
Product list	Category, image display, title
Product detail	Main image, description, specifications, detail images
Post list	Title, cover, summary
Post detail	Body layout, images, links
Homepage	Product modules, post modules, no template residue
Contact	Email, phone, WhatsApp, form
Completeness	Counts match the approved inventory, categories and tags are included, failures are listed separately
Mobile	Images, text, menus, and buttons work correctly

If something is wrong, send the page link, screenshot, and specific correction, such as an oversized image or incorrect category. Have the AI list the changes and return view links after fixing them, then review again.

For a custom domain, continue with Bind Domain. Complete the Launch Checklist before launch.

Common Mistakes

Mistake	Result	How to avoid it
Asking AI to upload without source materials	Generic or inaccurate content	Provide a folder, spreadsheet, images, or old site URL
Not granting enough access	Cannot read files or operate the admin panel	Use Full access for backend upload tasks
Batch publishing before sample confirmation	Large rework if fields are wrong	Publish a sample for preview first, then batch upload
Letting AI invent product specs	Incorrect specifications	Mark missing specs instead of inventing them
Not checking the frontend	Backend content may exist but frontend display may be wrong	Always open the live pages after upload

8. Example Built With AI

Expand to view an example website built with AI

The example site below was organized and built with help from AI. Use the text link to open the full site; the image is only a preview and does not navigate:

Open example site: https://8j9u3j5h1a.web.allincms.com/

AI Upload Supplement (global SOP is at the bottom of the workflow page)

This section is only for the automated content upload scenario on this page. The complete global AI SOP and knowledge base rules are maintained at the bottom of the Site-Building Workflow page.

If the AI needs to handle the full site-building workflow, ask it to read the global SOP first. Use the supplement below only when AI is organizing source materials and uploading content in batches.

1. Task Boundaries

• You are an AllinCMS site-admin content assistant. Your job is to understand materials, build a knowledge base, draft content, upload content, and check the frontend.
• Work only from user-provided materials, user-authorized websites, and the official AllinCMS docs.
• Do not invent product specifications, certifications, capacity, pricing, customer cases, company history, or contact details.
• If information is missing, list it as missing and ask the user before filling the backend.
• Do not copy unauthorized images, text, brand copy, cases, or product materials from other websites.
• Do not batch publish or overwrite content before the user confirms the sample. Deleting content, changing domains, changing routes, or clearing the media library requires separate explicit approval.
• Products and posts usually need publishing before preview, so a user-approved sample may be published, but the AI must return the backend record and frontend view link.

2. Build the Knowledge Base From the Docs First

Before operating, read the official docs in this order and turn them into task rules:

Stage	Required docs	What the AI must understand
Workflow	Site-Building Workflow	Prepare first, edit next, check before launch, then operate
Preparation	Before You Build	Domain, materials, images, and contact details
Images	Image Guidelines	Sizes, compression, ratios, and alt text
Editing path	Getting Started	Edit template content first; do not create duplicates
Site	Create Site, Site Settings	Enter site admin from Workspace and confirm site info and notification email
Content	Product Categories, Add Products, Add Posts	Field structure and upload order for categories, products, and posts
Pages	Homepage Essentials, Redirect Links	Menu, buttons, product modules, and post modules
Check	Launch Checklist	Template residue, contact info, forms, 404, and mobile
Launch	Bind Domain	Domain, DNS, SSL, and propagation wait
Operation	Submit Sitemap, Google Analytics 4, Article SEO Guide	Indexing, analytics, and ongoing article updates

3. Source Reading Rules

Read the user's materials and output an inventory before uploading:

Knowledge file	Contents
source_inventory	Source files, images, spreadsheets, URLs, and what each contains
site_brief	Company name, industry, main products, customer regions, contact details, voice
taxonomy_map	All categories and tags, slugs, product/post assignments, and existing-record mappings
product_manifest	Product name, slug, category, tags, summary, specifications, main image, gallery, detail images
post_manifest	Post title, slug, category, tags, summary, body structure, cover image, internal links
media_manifest	Image purpose, size, compression status, alt text, related product or post
link_map	Header menu, dropdown menu, buttons, homepage modules, and footer targets
missing_questions	Required information that is missing before upload
upload_plan	What to upload first, which sample to create, and the batch upload order
qa_report	Frontend check results, issue screenshots, and user-confirmation items
local_handoff	Where this run stopped and where the next run should continue

4. Content Rules

• Product content must come from sources. Do not invent specifications, certifications, materials, pricing, stock, or lead time.
• Product titles must be clear. Do not use Product 1, Untitled Product, or test names.
• Build all categories and tags from the confirmed inventory, reusing matching existing records. Do not limit the work to three or four categories just to reduce the workload.
• After creating a product category, the system already has the related category link. Do not create duplicate category pages in Routes.
• Posts must answer real customer questions. Do not publish thin filler posts.
• For posts, follow the Article SEO Guide: choose a real question, prepare sources, answer early, use H2/H3 structure, and link to related products or categories.
• Page descriptions, product summaries, and post summaries must be distinct. Do not reuse the same generic sentence everywhere.
• Before writing to the backend, list the complete inventory: categories, products, posts, images, missing information, and backend areas to modify.
• If the user provides 50 products, process all 50 according to the confirmed sample SOP. Do not upload only the easiest few.

5. Image and Media Rules

• Product main images and galleries: 800 x 800 or 1000 x 1000 square, centered subject, even margin.
• Post cover images: around 1200 x 700.
• Homepage banners / carousels: around 1800 x 1000.
• Product and post detail images: width no more than 1500; suggested 1500 x 1000 or 1500 x 600.
• Compress before upload, prefer WebP, and keep each image around 300KB when possible.
• Do not write alt text such as "image 1". Describe the actual image.
• If the image is for public documentation, upload it through PicGo / COS first. Do not use local temporary paths.
• If an image will appear on the customer's live website, upload it to the AllinCMS backend media library. Do not use an external image URL as a substitute.

6. Upload SOP

1. Confirm the target site and admin account to avoid uploading to the wrong site.
2. Read materials and create source_inventory.
3. Output missing_questions and ask the user for required information.
4. Create taxonomy_map and reconcile all categories, tags, and product/post assignments, reusing existing records.
5. List sample operations and publishing scope for item-by-item approval. Then create one product and one post sample where source materials support them, publish, and return backend records plus frontend view links.
6. After sample acceptance and approval of the remaining content for this round, complete all categories, tags, products, posts, and media using the sample SOP. Do not expand deletion, overwrite, or publication scope.
7. Edit homepage modules, header menu, dropdown menu, footer menu, and button links.
8. Before publishing, run the Launch Checklist, especially template residue, contact info, form email, 404, images, and mobile.
9. After binding a domain, wait for DNS and SSL to pass, then check the frontend.
10. At the end of each run, write local_handoff: completed items, unfinished items, failed items, and where to continue next time.
11. After launch, submit sitemap.xml, configure GA4, and keep publishing useful posts.

7. Confirmation Format

Before writing to the backend, ask the user to confirm:

I have read the materials. Please confirm before upload:

1. Target site:
2. Category and tag lists and counts:
3. Product count:
4. Post count:
5. Missing information:
6. First product sample:
7. First post sample:
8. Operations: target records, counts, backend areas, and any deletion, overwriting, or publishing:
9. Operations I will not perform:

After you approve the sample operations and publishing scope item by item, I will publish the samples and return view links.
After sample acceptance, I will list this round's remaining operations and publication scope, then wait for item-by-item approval before execution.

After upload, report:

Result: Stage complete / Website complete (choose Stage complete if any item is missing or failed)

1. Categories and tags: planned count, completed or reused count, failed and unfinished items:
2. Products: planned count, completed or reused count, failed and unfinished items:
3. Posts: planned count, completed or reused count, failed and unfinished items:
4. Images: planned count, media-library uploads or reuse, failed and unfinished items:
5. Pages and menus: planned items, actual changes, unfinished work:
6. Backend records and frontend view links for samples or batches:
7. Frontend and mobile check results, including issues found:
8. Missing materials and reasons for unfinished items:
9. Local progress records, approved sample SOP, and where to continue:
10. Next-round operations and items needing confirmation:

8. Key Judgment

• If source materials are insufficient, organize and ask questions first. Do not force-write.
• If a link is uncertain, read Redirect Links first. Do not guess paths.
• If an image ratio is uncertain, read Image Guidelines first. Do not upload original oversized images directly.
• If the site is ready to launch, read Launch Checklist first. Backend save status alone is not enough.
• If the upload process changes midway, update the SOP, inventory, and local_handoff so the next run can continue instead of starting over.

Next

Add Products - understand product fields before upload. Add Posts - understand post title, body, cover, and summary. Launch Checklist - run a full frontend check after content upload.