Make LLMs Actually Useful for Your Data.

Data Notes solve the context problem. Describe your database structure, table relationships, column meanings, and business logic in plain English. VerbaGPT automatically includes this context when generating queries.

Levels: Database, Table, Column, Value, and Sample questions. Non-prescriptive schema allows you to add exactly the context that matters.

Team benefit: Data admins curate once. All users benefit from accurate, context-aware queries automatically.

Most people don't know how to prompt LLMs effectively. Prompt Libraries democratize expertise by capturing proven, effective prompts in reusable collections.

Internal use: Share best practices org-wide. Marketing team's top 20 prompts → instant access for new hires.

Revenue opportunity: Consultancies can package domain expertise (e.g., "Healthcare Analytics Prompts" or "Financial Modeling Library") and license to clients.

Auto Rank: Type your question, VerbaGPT suggests relevant prompt templates from your libraries. One click to apply and customize.

Virgo can orchestrate multiple frontier models via modern providers, using one model for schema reasoning, another for code generation, and another for review — without changing your workflows.

You gain flexibility to experiment with new models and pricing, while keeping a single, consistent product experience for your teams.

Taurus is currently powered by Claude Code, with a model-agnostic architecture designed to support additional local agents as they become available.

With Private Inference, the model that writes and reviews Virgo's analysis code can run on your own machine instead — through Ollama, LM Studio, or any OpenAI-compatible engine.

Authentication & Access Control: JWT-based authentication, bcrypt-hashed passwords, and role-based permissions (user, admin, org) control access throughout the app. Sensitive operations verify ownership at the database level.

Isolated Code Execution: The code VerbaGPT writes for your analysis runs in a separate process that never holds the platform's own credentials, cannot reach the system database or other users' files, and is terminated automatically if it runs away. Your database credentials are never sent to an AI model.

Encryption at Rest: All data is encrypted at rest using AES-256 encryption, including user credentials, chat history, and datasource configurations. Database connection strings carry an additional layer of application-level encryption with key rotation.

Encryption in Transit: All connections use TLS 1.3 encryption (HTTPS) between your browser and VerbaGPT servers. Database connections use SSL/TLS encryption for secure communication.

Infrastructure: VerbaGPT runs on enterprise-grade managed cloud infrastructure with environment isolation, automated backups, and 99.9% uptime SLA.

Audit Logging: Enterprise-grade audit trail tracks all user queries, datasource access, and actions. Admin dashboard with stats, error monitoring, and CSV export. 30-day retention with automatic cleanup.

Privacy: VerbaGPT does not use your chat history or data to train AI models. Your data remains yours.

Read the full security overview →

VerbaGPT is Text-to-Python, not just Text-to-SQL. This unlocks capabilities impossible with SQL alone:

• Statistical modeling: Regression, time-series forecasting, hypothesis testing
• Machine learning: Classification, clustering, decision trees
• Advanced visualizations: Heatmaps, 3D plots, custom charts (matplotlib)
• Custom analysis: Combine multiple data sources, complex transformations

Example: "Create a predictive model for customer churn using decision trees. Show feature importance and confusion matrix." VerbaGPT generates Python code with pandas, scikit-learn, and matplotlib automatically.

The challenge: "What questions should I even ask?" New team members don't know what's possible. Footnotes solve this.

How it works: Colleagues add helpful prompts as "footnotes" tagged with context (e.g., "sales analysis" or "inventory forecasting"). VerbaGPT uses semantic search to surface relevant examples when you need them.

Organization-wide learning: Junior analysts benefit from senior colleagues' expertise. Tribal knowledge becomes institutional knowledge.

Built for consultancies and agencies: Publish a database, invite your client's email, and they can ask it questions in plain English — no VerbaGPT subscription, no per-seat fee on their end. It's covered by your organization's own credit pool.

Scoped access: Clients see only the dataset you shared with them, not your other data, prompts, or team resources. Revoke access instantly at any time.

Spend control: Set a credit cap per client and an aggregate cap across all your clients combined, so one invite can never run away with your budget.

VerbaGPT builds a private understanding of your data as your team works — the real column names, how your codes and dates are stored, the queries that already worked, the corrections you've made.

Correct an answer once and it offers to save the fix as a reusable rule. Thumbs-up a result and it reuses that approach on similar questions later, citing it in Sources.

When a question keeps tripping it up on a database — a wrong column name, a status stored as A instead of active, an odd date format — a daily pass drafts improvements to that database's Data Notes for you to review and approve.

Upload a data dictionary or spec doc and it extracts column notes, table relationships, and guidance as drafts.

You stay in control: every learned item is scoped to you or your organization, shown for review, and reversible. Nothing is used to train shared models.