Unlock the Power of AI in Your Workflows: Introducing AI Actions Pipelines Channel

A few months back, I wrote about the OpenAI channel in Pipelines and the benefit it could have to expand on the use cases that can be solved with Quickbase.

Now, I'm writing again, but for something even more exciting - AI Actions.

With this new channel, you will be able to safely and securely connect with an AI provided by Quickbase and send any information (yes, including files) along with instructions for the AI to perform. No API keys and no AI model training.

AI Actions allows Quickbase builders to automate complex text and data tasks — from summarizing emails to analyzing uploaded files — without leaving Pipelines.

What is AI Actions?

To put it simply, a new pipelines channel, but able to solve countless problems.

The channel has a single step called "Custom Action" and it allows the builder to pass any information and instructions to the AI and request from the AI to return its answer in a structed manner, so it can easily be referenced in following steps.

Step parameters:

• Response Format - Text (default) or JSON
• AI Persona - instructions for the AI to carry out
• Write your request - information that you'd like the AI to perform the instructions on
• File URL - Quickbase file attachment

 

See it in Action

I've picked my first use case to be a very simple one - extracting specific information from incoming emails. You can watch this next part as well here.

This is very common for any business that has customer service.

The pipeline looks like this (YAML at the bottom):

1. A new email is received.
2. The AI extracts key information from the email based on the pre-defined instructions.
3. The information from the AI is recorded in a table.

The interesting part is splitting the response from the AI, so it can be easily recorded in different fields. This is achieved by providing a JSON schema for the output for the AI to fill. The JSON schema tells the AI how to organize its answer — like a template it must fill in.

JSON Schema:

{
  "type": "object",
  "properties": {
    "Summary": {
      "type": "string"
    },
    "Urgency": {
      "type": "string",
      "enum": ["Low", "Medium", "High"]
    },
    "Customer": {
      "type": "string"
    },
    "Case_id": {
      "type": "string"
    }
  }
}

Limitations

• Available for Business and Enterprise plans.
• Daily limit on amount of information sent to and received from the AI.
• Provided files need to be stored in Quickbase and up to 2 MB.
• The AI cannot recognize objects on images, but can extract text from them.
• PII data is not processed.

To read the full list of limitations, check here.

Conclusion

The introduction of AI Actions opens up a whole new world of possibilities for Quickbase builders. From summarizing customer emails to analyzing uploaded files or automating data entry — the potential use cases are endless.

If you’re already using Pipelines, give AI Actions a try and share what you build!

I’ll be exploring more advanced examples in future posts, so stay tuned.

 

Example pipeline YAML

# Email summary
#

# Account slugs:
#  - microsoft-outlook[REDACTED]: Georgi Peev (gpeev@quickbase.com)
#    <None>
#  - quickbase[REDACTED]: Realm Default Account <None>
---
- META:
    name: Email summary
    enabled: false
- TRIGGER microsoft-outlook[REDACTED] email on_create -> a:
    FILTERS:
    - AND:
      - subject contains Test Case
    name: New Support Case
- ACTION qb-ai-actions custom_action create -> b:
    inputs-meta:
      json_schema: "{\n  \"type\": \"object\",\n  \"properties\": {\n    \"Summary\"\
        : {\n      \"type\": \"string\"\n    },\n    \"Urgency\": {\n      \"type\"\
        : \"string\",\n      \"enum\": [\"Low\", \"Medium\", \"High\"]\n    },\n \
        \   \"Customer\": {\n      \"type\": \"string\"\n    },\n    \"Case_id\":\
        \ {\n      \"type\": \"string\"\n    }\n  }\n}"
      response_format: json_schema
    inputs:
      system_message: 'You are an assistant that checks incoming emails from customers.

        Your job is to extract key information from the email in a JSON.

        - What is the urgency according to the tone of the email and severity of the
        issue?

        - Who is the customer? If you cannot find a company name, use the sender email
        domain.

        - Is there a case ID in the email? If not, write "No Case ID"

        - Write a short summary of the email with no more than 100 words.'
      user_message: 'Email content:

        From: {{a.from_address.email_address.address}}

        -----

        Subject: {{a.subject}}

        ------

        {{a.body}}'
    name: Extract Info
- ACTION quickbase[REDACTED] record create -> c:
    inputs-meta:
      export_fields: '"original email, case ID, customer, summary, urgency" <10, 8,
        7, 6, 9>'
      table: '"AI Actions - Georgi Peev: Email summary" <bviknueix>'
    inputs:
      case_id: '{{b.output_json.Case_id}}'
      customer: '{{b.output_json.Customer}}'
      original_email: 'From: {{a.from_address.email_address.address}}

        -----

        Subject: {{a.subject}}

        ------

        {{a.body}}'
      summary: '{{b.output_json.Summary}}'
      urgency: '{{b.output_json.Urgency}}'
    name: Create Case in QB
...