The chatbot allows users to interact with collections using a natural language interface, making it easier to query data and extract insights without needing deep technical knowledge. Here’s a breakdown of how to use it effectively:

Settings

Before using the chatbot, you can configure the following options:

1. Collection types:

• Choose which collections the chatbot will search (public collections, collections shared with you, your private collections)

2. Web Search:

• Enable or disable web search to supplement chatbot responses when data isn’t available in collections.

3. Proof Execution:

• Decide whether to generate a proof of the query result.
• Keep in mind that enabling this feature increases token consumption, as it involves additional computational steps for ZK-proving.

Asking Questions

Once your settings are configured, you can ask the chatbot any question. The system will attempt to retrieve relevant information from the available collections and provide an answer.

Example Questions:

• "What is the total revenue for Q3 2023?"
• "How many basketball games were played last season?"
• "What is the average score of Team A?"

For the alpha version, Provably includes public collections with basketball-related data to help you test this functionality. To maximize success, review the available collections and their data beforehand to ensure your questions can be answered.

Current Limitations

While the chatbot is powerful, there are important constraints to keep in mind:

1. Aggregate Functionality Only:

The chatbot only supports the following SQL aggregate functions:

• SUM, COUNT, AVG, VARIANCE, STDDEV, MEDIAN, MIN, MAX

Supported questions include:

• "What is the sum of all sales?"
• "How many players are in the dataset?"
• "What is the maximum score recorded in a game?"

2. Unsupported Query Types:

The chatbot cannot answer questions that require:

• Individual Record Lookups (e.g "Which player scored the most points in a single game?")
• Ranking and Sorting (e.g "Show me the top 10 highest-scoring players.")
• Conditional Grouping or Multi-Group Aggregates (e.g "What are the average sales grouped by product type?")

3. Complex Joins or Relationships:

Queries requiring joins or relationships between multiple tables are not supported.

4. Web Search Limitations:

• Web search results are not ZK-proven.
• The chatbot will indicate when responses include web-based data.

5. ZK-Proving Constraints:

Zero-knowledge proving imposes additional limitations:

• Recursive queries and iterative calculations are unsupported.
• Dynamic or parameterised queries cannot be executed.

Alpha Testing Data

For testing, public basketball datasets are available. Explore these collections to understand their structure and content before asking questions. This helps ensure the chatbot can generate accurate responses.

By understanding these limitations and leveraging the supported query types, you can make the most of Provably’s chatbot while adhering to the system’s current constraints. Future updates aim to address these limitations and expand functionality.