SnapGPT

Overview

SnapGPT is an AI-driven platform that helps you generate your pipelines from simple text prompts.


SnapGPT panel in Designer

Features

You can use SnapGPT to:
  • Build pipelines using simple prompts
  • Describe pipelines
  • Analyze pipelines
  • Ask anything about the SnapLogic Intelligent Integration Platform (IIP)
  • Generate SnapLogic expressions
  • Create SQL queries
  • Detach the panel and move it anywhere on the Designer canvas. You can drag and resize the panel to see chat content clearer.
In upcoming versions, you will be able to:
  • Build complex pipelines
  • Configure your accounts and pipelines with the Config Co-pilot
  • Generate better SQL queries based on your schema
  • Integrate directly with your LLM provider
Tip: For early access to upcoming features, check out SnapLabs.

Change in behavior

Because SnapGPT already provides a contextual generate expressions feature in Snaps, the Generate expression functionality in the SnapGPT panel is removed. Instead, use the generate expression capabilities in the Expression field of a Snap Settings dialog.

Retrieval Augmented Generation

In the March 2024 release, SnapGPT introduced Retrieval Augmented Generation (RAG) as a way to improve results based on existing artifacts in your environment (Org). RAG is an implementation that helps power accurate results based on what the user is asking for. Because other LLMs do not have context before 2022, RAG adds those extra contexts from your knowledge archive or personal databases. RAG also decreases the chance of LLM hallucinations when the most accurate answer is unavailable.

RAG settings matrix

RAG settings are set by the Org admin in Admin Manager > SnapGPT Settings. The Org settings determine the settings available to the user.

Users can also specify the RAG settings in their interaction with SnapGPT in the Use my existing pipelines for enhanced pipeline generation dropdown.

The following table describes the relationship between Org and user settings:

Org Setting User Setting Result
Auto Auto

When Auto is selected, pipelines are recommended by SnapGPT. Suggests 5 pipelines for you to choose from, to generate a new pipeline.
Manual or Disable You can also choose either to select the recommended pipelines manually or to disable the RAG settings entirely.
Manual Manual or Disable You can also choose either to select the recommended pipelines manually or to disable the RAG settings entirely.
Disable N/A RAG is disabled.

Select Recommendations using RAG tools

After you ask SnapGPT an initial question, it generates a pipeline. When the pipeline is generated, SnapGPT displays a dialog for you to select the pipeline to be used as context.


SnapGPT RAG window

  1. In the Relevant pipelines list, you can view the name and path of the pipeline at the top. The preview provides the following actions:
    • Click to open the pipeline in Designer.
    • Select to view a preview of the pipeline shape.
    • View a list of pipelines from the user’s organization, and select a pipeline from this list.
  2. Click Add to move to Pipelines to be used window.
Note: If set to Auto, the process of selecting the pipeline is completed by SnapGPT.

Effective prompts

The usefulness of the results you get from SnapGPT depends on the quality of the prompts you enter. The best prompts are:
  • Clear: Be precise to help SnapGPT determine the most suitable response.
  • Concise: Be brief to reduce latency in responses.
  • Consistent: Be consistent to help SnapGPT learn quickly and avoid variability in the results.

You must be very specific in your request to get the best possible experience from SnapGPT.

Keep refining your questions to get the best results from SnapGPT.
  • Good: how do i schedule a pipeline?
  • Better: how do i schedule a pipeline to run on a daily basis?
  • Best: how do i schedule a pipeline to run every day at 9AM UTC?

Dos

  • Be specific in your prompt. For example, specify the endpoints you want to integrate (such as Salesforce, Snowflake, Google BigQuery, AWS S3, or REST API). Specify the data you want to extract or transform, such as a Salesforce Opportunity.
  • Use the prompts suggested by SnapGPT. Replace the highlighted portion with your specific use case.

Don'ts

  • Do not ask SnapGPT general questions such as What is the capital of Florida?
  • Do not include sensitive data such as account credentials in your prompt. You can enter the account details when you configure the account.

Example prompts

  • Create a SnapLogic pipeline to join tables t1 and t2 and insert the output in table t3.
  • Generate and save a SnapLogic pipeline to create a Snaplex. Open the REST Post Snap to configure.
  • Create a SnapLogic pipeline to download a CSV file from S3 and query the CSV file for records that have the zip code 78660.
  • Generate a pipeline with a JSON Generator that contains 10 sample Salesforce lead objects.
  • Read a file in CSV format and write to Google Sheets.
  • Get Salesforce opportunities and write to BigQuery.
  • Get exchange rates from the European Central Bank and write the output to a JSON file.
  • Use an API to compute carbon emissions based on 50 kWh electricity consumption in FL, USA.

For more examples, refer to the SnapGPT Prompt Catalog

Known issues

  • You might experience significant latency (up to one minute) or timeout issues, depending on the complexity of the request and the capacity of the LLM provider.
  • As with all large language models, SnapGPT can occasionally be unreliable or inconsistent in its responses.
Note: SnapGPT is powered by publicly-available large language models (LLMs), so the prompts you enter query the entire dataset of the LLMs and is not limited to specific SnapLogic data. Generic parts of prompts are sent anonymously to the supported LLMs.