Skip to main content

Chat GPT Integration

We have an integration to Chat-GPT in our platform, which is implemented for search purpose.

Demo​

For complete step-by-step instruction see Quick Start Demo

Create config​

  1. Go to Quasiris Search Cloud
  2. Enter 'Config' in the top left search input
  3. Click on 'Config'
  4. Create a config with name 'chat-gpt-integration-config'
  5. In tab 'Config' post a json 
    {
      "type": "searchResultRefinement",
      "searchResultRefinement": {
         "requestTemplate": "Title is '${(searchResults['chat-gpt-quick-start-demo'].documents[0].document.title)!\"\"}'",
         "appendToSearchContextKey": "biking",
         "contextString": "This is a chat for the application, where I would enter a query, and you would answer the question from this query. Possible answers are only yes, no, unknown. Don't add any comments or symbols. ONLY 'yes', 'no', 'unknown'. \nWe are at biking shop, I want to know if the title user enter somehow connected to biking topic."
      }
    }
  6. Replace 'chat-gpt-quick-start-demo' with the name of your search config, where you want to use it
  7. Click 'Save'

Json structure​

{
  "type": "searchResultRefinement",
  "searchResultRefinement": {
     "requestTemplate": "Title is '${(searchResults['chat-gpt-quick-start-demo'].documents[0].document.title)!\"\"}'",
     "appendToSearchContextKey": "biking",
     "contextString": "This is a chat for the application, where I would enter a query, and you would answer the question from this query. Possible answers are only yes, no, unknown. Don't add any comments or symbols. ONLY 'yes', 'no', 'unknown'. \nWe are at biking shop, I want to know if the title user enter somehow connected to biking topic."
  }
}
  • type - must be searchResultRefinement to use it with ChatGPT filter
  • searchResultRefinement.requestTemplate - freemarker template, which would be used to send requests to ChatGPT
  • searchResultRefinement.appendToSearchContextKey - an alphanumeric value to determine json key where we put a ChatGPT answer in search response
  • searchResultRefinement.contextString - a context for ChatGPT. A string where you would enter the instructions for chat GPT and preconfigure what it needs to do.
  1. Go to your search config
  2. Go to 'Pipeline' tab
  3. Choose 'Custom' pipeline type
  4. Post yaml
!!com.quasiris.qsf.pipeline.Pipeline
id: search-pipeline
timeout: 4000
filterList:
  - !!com.quasiris.qsf.pipeline.filter.DebugFilter
    active: true
    id: search-pipeline.DebugFilter
  - !!com.quasiris.qsc.search.service.query.QscSearchQueryFilter
    active: true
    id: search-pipeline.QscSearchQueryFilter
  - !!com.quasiris.qsc.search.service.rules.QueryRuleFilter
    active: true
    baseUrl: {{ searchIndexBaseUrl }}
    env: {{ env }}
    id: queryRuleFilter
  - !!com.quasiris.qsf.pipeline.filter.elastic.ElasticFilter
    active: true
    baseUrl: {{searchIndexBaseUrl}}
    id: searchElasticFilter
    resultSetId: {{searchCode}}
    queryTransformer: !!com.quasiris.qsc.search.service.QscSearchQueryTransformer
      profile: {{profile}}
    searchResultTransformer: !!com.quasiris.qsc.search.service.QscResultTransformer
    elasticSearchClient: !!com.quasiris.qsf.commons.elasticsearch.client.ElasticSearchClient
      useApacheClient: false
  - !!com.quasiris.qsf.pipeline.filter.ConditionFilter
    active: true
    condition: !!com.quasiris.qsf.pipeline.filter.conditions.EmptyResultCondition
      resultSetId: {{searchCode}}
    id: conditional.isSpellcheck
    pipeline:
      filterList:
        - !!com.quasiris.qsf.pipeline.filter.TokenizerFilter
          active: true
          id: qsc-spellcheck.TokenizerFilter
        - !!com.quasiris.qsf.pipeline.filter.elastic.SpellCheckElasticFilter
          active: true
          baseUrl: {{searchIndexBaseUrl}}-spellcheck
          id: qsc-spellcheck.SpellCheckElasticFilter
          maxTokenLenght: 10
          minTokenLenght: 4
          minTokenWeight: 1
          sentenceScoringEnabled: false
          restartPipelineId: search-pipeline
      id: qsc-spellcheck
      timeout: 1000
  - !!com.quasiris.qsf.pipeline.filter.QSFQLResponseRefinementFilter
    active: true
    id: search-pipeline.QSFQLResponseRefinementFilter
    resultId: {{searchCode}}
  - !!com.quasiris.qsc.chatgpt.filter.ChatGPTFilter
    active: true
    apiKey: "chat-gpt-integration-config"
    modelId: "gpt-3.5-turbo"
    chatGptConfigCode: "chat-gpt-config"
    resultSetId: {{searchCode}}
    maxTokens: 10
    timeoutMills: 8000
    temperature: 0
  - !!com.quasiris.qsf.pipeline.filter.elastic.ElasticTrackingFilter
    active: true
    baseUrl: {{trackingIndexBaseUrl}}
    customParameter:
      app: search
      searchCode: {{searchCode}}
      tenantCode: {{tenant}}
      env: {{env}}
    id: search-pipeline.ElasticTrackingFilter
    idFieldName: id
    resultSetId: {{searchCode}}
    rotation: none
    trackingId: null
  1. Click 'Save'

Yaml structure​

- !!com.quasiris.qsc.chatgpt.filter.ChatGPTFilter
 active: true
 apiKey: "chat-gpt-api-key"
 modelId: "gpt-3.5-turbo"
 chatGptConfigCode: "chat-gpt-config"
 resultSetId: {{searchCode}}
 maxTokens: 10
 timeoutMills: 8000
 temperature: 0
  • active - specify if filter must be executed
  • apiKey - api key for ChatGPT
  • modelId - model id of ChatGPT
  • chatGptConfigCode - the config code, which you created in previous steps
  • resultSetId - result set id, where you want to put chat gpt response
  • maxTokens - maxTokens configurations for ChatGPT
  • timeoutMills - timeout for the operation
  • temperature - temperature for ChatGPT requests (0 - the most precise, 1 - the most unpredictable)

Now you can make search with the search config and find chat gpt response in context fields. For example if you have search config with code 'demo', the json path to retrieve the context would be $.result.demo.context