• English

      • AI Model

      Setup > AI Model is where you compare models before sending real shoppers to the widget.

      What this page is for

      Use it to:

      • Select the model HeiChat should use
      • Compare token multipliers
      • Test responses in the admin first
      • Decide whether quality gains justify higher cost

      A practical testing workflow

      1. Define test questions

      Prepare real questions from your store, such as:

      • Product discovery
      • Shipping and returns
      • Order-related scenarios
      • Edge cases that are important for your team

      2. Test more than one model

      Try:

      • Basic models for cost efficiency
      • Advanced or premium models for higher accuracy

      Compare:

      • Accuracy
      • Product recommendation quality
      • Tone
      • Token cost

      3. Fix setup issues before blaming the model

      If answers are weak:

      • Check Learned Products if product recommendations are missing
      • Check Knowledge Base if store-specific information is missing

      Many quality issues come from incomplete store data, not only from model choice.

      4. Roll out only after admin testing

      For cautious launches:

      1. Keep storefront activation off at first.
      2. Test models and refine knowledge.
      3. Activate the widget only after results are acceptable.

      Model cost logic

      HeiChat labels models by tier and token multiplier relative to the baseline model.

      As a rule:

      • Lower multipliers are cheaper
      • Higher multipliers often give better reasoning

      Choose based on your actual traffic pattern, not only on lab-style comparisons.

      Custom OpenAI API key

      The Custom OpenAI API Key is available on Pro Max. It is a continuity fallback: HeiChat uses included plan tokens first, then any Extra Tokens. Only after both balances are exhausted does it automatically use your saved OpenAI key.

      Make sure the key is valid and that the connected OpenAI account has sufficient balance. Usage through that key is billed by OpenAI. Reply style, language, length, emojis, and instructions are configured separately in Advanced Settings > AI Response.

      Reducing token usage

      Two independent choices help control cost:

      • Choose a Basic (1x) model when lower cost matters more than maximum reasoning quality.
      • Put repetitive high-volume questions into Quick Buttons > Q&A. These replies use your configured answer directly and do not perform additional knowledge retrieval for that turn.