WaaS Help Center

AI Assistant

The AI Assistant is built directly into your WaaS backend. You chat with it in natural language, and it helps you create, translate, rewrite, and analyze content — without ever leaving the editor.

This page describes what you can do as an editor or admin of your WaaS instance. Topics like provider setup, API keys, or model selection are handled by Blackpattern — that’s part of your service contract and we take care of it.


Where do I find the AI Assistant?

In every WaaS backend, you’ll find the AI Assistant button on the right side of the sidebar. One click opens the chat panel on the right edge of the screen. The panel stays open while you navigate through the backend — so you can work in your content and chat with the assistant in parallel.

Which AI model is used?

Your WaaS is configured with GPT 5.4 by default. Model selection is handled by Blackpattern — you don’t have to worry about API keys, providers, or model updates. If you need a different model (for example, because you frequently analyze large documents), talk to us.


What can the AI Assistant do for me?

Create and edit content

The assistant has direct access to your collections (e.g. Pages, Posts, Services). You can ask it to:

  • create new entries (“Create a new blog post with title … and status draft”)
  • find existing items (“Show me all articles from this month without a featured image”)
  • update specific fields (“Update this page’s meta description to …”)

Before the assistant actually writes anything, it asks you for confirmation — see the Tool approvals section below.

Analyze files

You can drag and drop files into the chat or pick them from your File Library. Supported formats:

  • Images (JPEG, PNG, GIF, WebP) — e.g. “Describe the image for the alt text” or “Which colors dominate?”
  • Documents (PDF, text) — e.g. “Summarize this PDF in 5 bullet points”
  • Video (MP4)

Maximum file size: 50 MB. Important: uploaded files go to the AI provider, but they are not automatically added to the media library in your WaaS backend.

Reusable prompts (brand voice)

If your WaaS instance uses the Prompts collection, you can attach saved prompts like “Write in our brand voice” or “Generate an SEO description” with a single click. Prompts can contain variables like {{topic}} — the assistant will then ask for them specifically.

If you’d like to use this feature and it’s not set up for you yet, get in touch with us.


Attaching context — the most important lever

Via the + button next to the input field, you can give the assistant specific context:

  • Prompts — saved templates
  • Items — up to 10 records from any collections (e.g. two products to compare)
  • Files — upload from your computer or pick from the File Library

You can attach up to 10 context elements per message. What you attach is sent as a snapshot at the moment you submit — later changes to the original are not visible.

Rule of thumb: The more specific the context, the better the result. “Write a text” → mediocre. “Write a text in the style of this attached article for this attached product page” → noticeably better.


Tool approvals (security model)

The assistant can’t simply create or delete things without asking you. Every time it wants to use a tool (e.g. writing a record), an approval dialog appears with three options:

  • Approve — run once
  • Reject (or Esc) — cancel the action
  • Always allow — run this tool in the future without asking

The settings are stored locally in your browser. They only apply to you, not to other editors.

Recommendation for editors:

  • Read operations (searching content, viewing the schema) can safely be set to “Always allow”.
  • Write and delete operations should be left on “Approval required” — otherwise the assistant might, in rare cases, overwrite something unintentionally.

Via the cog icon in the chat header, you can decide per tool how it should be handled.


Conversations and privacy

What is stored where?

  • Chat histories live exclusively in your local browser (Local Storage). They are not stored on the WaaS servers and are not shared between devices or users. Clear your browser = history gone.
  • API keys are stored encrypted in the database and are only visible to Blackpattern (admin).
  • Your inputs, schema info, and attached content are transmitted to the AI provider with every request (see below).

Where does the data go?

Requests are sent to the AI provider for GPT 5.4. This includes:

  • the text you enter,
  • the content of the items and files you attach.

⚠️ Please note: Don’t send third-party personal data, credentials, or confidential business information to the assistant if you’re unsure about the provider’s processing policies. When in doubt: before attaching a file, take a moment to consider whether the content is suitable for external processing.

What can the assistant “see”?

The assistant always operates with your permissions. If you don’t have access to a particular collection, neither does the assistant. This rules out anyone using the chat to get at content they wouldn’t normally be allowed to see.


What the assistant doesn’t do

To avoid any confusion — these things do not run through the chat, but through Blackpattern:

  • Model selection and provider configuration — we set that up.
  • Schema changes (new collections, new fields, relationships) — we handle that as part of your service contract.
  • Creating or changing flows and automations — we do that.
  • Larger bulk operations (e.g. updating 5,000 items) — the chat is the wrong tool for this. Talk to us, we’ll handle it via import/export or a script.
  • Sensitive production operations like mass deletions — please don’t do those without checking in first.

If you have the feeling that something on this list comes up regularly: good chance we can build it into your instance as a reusable prompt, flow, or function.


Tips for good results

Be specific. “Create a new article in posts with title ‘XYZ’, status draft, category ‘Tutorials’” produces noticeably better results than “Make a new post.”

Tasks in steps. For more complex tasks: first search, then display, then change. “Show me all products without a description” → “Update product X with this description …”.

Start fresh for new topics. Long conversations can lose the thread. If the assistant starts to seem confused or answers become off-target: clear the chat via the refresh icon and start fresh.

Context beats explaining. Instead of typing out the content of a page, attach it as an item. Instead of describing the content of an image, upload the image.

Quickly review before confirming a write action. The approval dialog shows you exactly what’s going to happen. A glance at the fields, then approve.


Example prompts for everyday WaaS work

Content editing

  • “Write three suggestions for an SEO description for this page (max. 155 characters).”
  • “Create a draft for a blog post on the topic … — status draft, category News.”
  • “Find all pages where the meta title is empty.”

Images & media

  • “Describe the attached image for an accessible alt text in German, max. 125 characters.”
  • “What mood does this photo convey? Suggest two suitable captions.”

Documents

  • “Summarize this PDF in a press-release-ready short version (max. 80 words).”
  • “Extract all dates and deadlines from this PDF as a list.”

Data research

  • “How many posts were published last month, grouped by category?”
  • “Show me the 10 most recently modified pages.”

Frequently asked questions

Are my chats archived? No. Chats live exclusively in your browser. If you close the browser or clear the cache, the history is gone. That’s intentional — nothing from the chat is stored long-term.

Can the assistant delete content without me noticing? No. Write and delete operations require your explicit approval by default. You have to actively click “Approve”.

What happens if I see an error in the chat? Often a click on the Retry icon helps. If the problem persists, clear the chat and try again. If it still persists: let us know.

Can I switch the model myself? No — the model and provider are configured centrally by Blackpattern. If you feel that the current model isn’t a good fit for your use case, talk to us. We’ll evaluate alternatives and switch if needed.


Service boundaries: Blackpattern ↔ Customer

AreaYou as customerBlackpattern
AI model, provider, API keys
Tool approvals (your own browser)
Reporting a model-switch request
Implementing a model switch
Creating / editing / translating content
Schema changes (new collections / fields)
Setting up reusable prompts
Using prompts in everyday work
Larger bulk operationsRequest