Skip to main content
Version: 3.5

AI Assistant Overview

CelerData offers an AI assistant - Agent Fawkes to help you understand and manage your data with ease. Whether you need answers to documentation questions, assistance with reading the database schema and task status, or help with SQL tasks (such as generating, fixing, rewriting, optimizing, or explaining queries), Agent Fawkes is ready to support you. It also auto-suggests charts to help visualize your query results.

Agent Fawkes operates as a set of cooperative services, most of which run within your own cloud region to ensure data locality. While Large Language Model (LLM) calls and documentation retrieval happen through a central cluster, your row-level user data never leaves your VPC, ensuring the highest level of security and privacy for your sensitive information.

Key Features and Capabilities​

  • Q&A Chatbot

    Agent Fawkes can respond to questions related to cluster use by referencing official documentation, ensuring users receive accurate and relevant information.

  • SQL Generation, Optimization, and Interpretation

    • Agent Fawkes can automatically generate SQL statements based on natural language inputs and existing schema information.
    • It can also correct, optimize, and rewrite SQL queries to improve performance.
    • It can interpret SQL queries, providing natural language explanations of their purpose and function.

  • Schema Inspection

    The Agent Fawkes retrieves and provides detailed metadata about the database schema, helping users better understand the structure of their data.

  • Task Monitoring

    Agent Fawkes can monitor the status of ongoing tasks and load jobs, providing real-time updates and feedback on their progress.

  • Chart Generation Suggestions

    Based on the query results, Agent Fawkes can auto-suggest relevant charts to visualize the data, improving insight and decision-making.

important
  • Cluster metadata (database schema or task status) is read with user consent collected by authorization prompts. You can choose Allow once or Allow for this session. The choice is remembered at thread granularity.
  • CelerData will not read row-level data from your cluster.

Data Privacy and Security​

  • Row Data Never Reaches the LLM

    All data processed by Agent Fawkes, including chart generation, is limited to column-level information (such as names, types, and statistics). Sensitive data is protected by a 3-layer sanitizer that masks PII, applies differential privacy techniques, and redacts sensitive column names.

  • Credentials Remain Secure

    Database credentials are stored in the user-center and never exposed to Agent Fawkes. The ai-agent uses an opaque driver ID, and actual database login details remain in a secure, in-memory connection pool.

  • Encrypted Conversation History

    Conversation data is encrypted at rest using KMS envelope encryption, with encryption keys rotated every 30 minutes to enhance security.

  • Cluster Data Locality

    Agent Fawkes operates within the user’s VPC, ensuring that all cluster data stays within the private environment, never being exposed externally.

  • Explicit User Consent

    The system prompts the user for permission before accessing any cluster data, either for a single session or for a one-time operation, ensuring full control over what information is shared.