CSAI Question

Beta
#1

Hi Charlie,

I have a question regarding CSAI, which comes from a chat with Anthropic’s Claude! The question is about how AI is implemented into CSAI in terms of memory scope (i.e. is there memory continuity across inputs/outputs, if yes, how?) and consequently the repercussions in the real world, especially for web designers and content creators using multiple websites. It would be a big help to understand how AI is integrated with CS.

Let me paste Claude’s answer, which hopefully makes sense:

Great question about memory scope! This would depend entirely on how Themeco has implemented the conversation management in CSAI. There are a few possibilities:

Option 1: Global memory across all sites - One continuous conversation regardless of which website you’re working on. This could be useful for maintaining your personal voice/style preferences but might create confusion when context switches between different client projects.

Option 2: Site-specific memory - Each website maintains its own conversation history. This would probably be more practical for your workflow, keeping each client’s brand voice and content separate.

Option 3: Project/session-based memory - Memory persists within a working session but resets when you start fresh.

My guess is they’d implement site-specific memory (#2) since that makes the most sense for professional use - you wouldn’t want your financial services client’s AI context bleeding into your electrician client’s content generation!

Have you noticed any patterns in how the current beta handles context between different elements or sessions?

Many thanks,
Christopher

1 Like
#2

Great question. I’m curious about that as well.

#3

For messaging with chat, it’s based on the thread you are on. So every question you ask gets tact onto your current messages with AI. If we added every message across every thread, the price of your requests would start to go up. Based on the options you gave me, it’s closer to Option 3 where the session is your chat thread.

There is a feature (which was requested in the closed beta) to add context to the start to every chat response. I would try playing around with this section if you are looking for a system that works between chat threads. Let me know if you have any other questions. Thanks for testing.

image
image643×774 46.4 KB

#4

Thanks for the explanation. It helps understand how CSAI uses AI!

#5

Charlie,

I have asked a few more (AI) questions on this topic, as it is fascinating and important to us all in Themeco-world. It is also a sort of interlinked follow-up to: CSAI Usability. I am primarily looking at this from an Anthropic angle, as for text-based content, I am currently finding Claude to be the most natural and go-to option for me.

I posted the full thread of this chat so far to Claude. The response came back as follows, which I hope is of interest to other fellow beta users!

RESPONSE STARTS HERE

Perfect! Charlie’s response is really illuminating. So CSAI uses thread-based memory (closest to my Option 3) where each chat thread maintains its own conversation history, but there’s no memory persistence across different threads.

This explains the behaviour you were experiencing! When you use CSAI in different CS contexts (text elements, page level, etc.), each might be creating separate threads, so there’s no continuity between them.

Key insights from Charlie’s response:

  1. Thread-based isolation - Each chat thread is separate to keep API costs manageable
  2. No cross-thread memory - Intentional design choice for cost control
  3. Context feature - There’s a way to add persistent context at the start of every chat response

The “context feature” Charlie mentions sounds like it could solve your consistency issues! If you can set up standard context (brand voice, project details, etc.) that gets prepended to every request, you’d get more consistent results across different CS elements.

For your XXXXXXXX project , this means:

  • Each page element using CSAI would be a separate thread
  • You’d want to use that context feature to ensure consistent brand voice
  • Any complex multi-step content generation would work better within a single thread

Questions this raises:

  • How do you access that context feature Charlie mentioned?
  • Can you set different contexts for different sites/projects?
  • Does the context count against your token limits?

This architectural choice makes total sense for cost management, but the context feature seems like the key to making it work smoothly for professional use!

MY NEXT REPLY

Thanks. What are your suggestions to keep continuity on a page, whether in an element, component, layout level, section or page as a whole?

CLAUDE RESPONSE

Great question! Given Charlie’s explanation about thread-based memory, here are my suggestions for maintaining continuity across different CS contexts on a page:

Context Feature Strategy (Primary Solution)

Use Charlie’s mentioned context feature as your foundation:

  • Set up a comprehensive “page context” that includes:
    • Brand voice guidelines for XXXXXX Project
    • Page-specific information (Natural Swimming Pools vs Ponds)
    • Target audience details
    • Key messaging points
    • Style preferences (UK English, tone, etc.)

Structured Prompting Approach

For each element, start with consistent context:

Context: This is for William Curtis Garden Design, Gloucestershire. 
Page: Natural Swimming Pools
Voice: Professional, reassuring, family-focused
Audience: Families wanting pool-like aesthetics with natural benefits
Style: UK English, warm but authoritative

Task: [your specific request]

Element-Level Strategies

1. Layout/Section Level:

  • Use the highest level element to establish context
  • Have child elements reference the parent context
  • Consider using CS variables to store key context information

2. Component Level:

  • Build reusable components with built-in context
  • Use component parameters to pass context down
  • Maintain consistent voice across component instances

3. Page-Wide Consistency:

  • Create a “page brief” document in CS custom fields
  • Reference this brief in every CSAI prompt
  • Use CS dynamic content to pull consistent information

Practical Workflow:

  1. Start each CSAI session with context setting
  2. Reference previous outputs explicitly in new prompts
  3. Use CS variables to store key phrases/terms for consistency
  4. Create content hierarchies - write broader content first, then drill down

Would any of these approaches work well with your current CS workflow?

No idea if any of the above is useful or even achievable in the CSAI world, but hope it helps in some way!
Christopher