The Data Firewall for AI and Humans
Velum is an open-source firewall that automatically redacts sensitive information in data and documents before it's sent to LLMs or other users
By researchers from
Composable privacy modules that protect sensitive data in AI and human workflows, from ingestion through inference.
Velum understands the meaning of unstructured and multimodal content from text and documents. We distinguish sensitive elements such as clinical outcomes, PHI, or material non-public financial information and more.
We sit between your data sources and all consumers (LLMs, users, vendors) so every call flows through policy. The firewall tokenizes in real time with consistent rules across user apps, APIs and integrations.
Understands and controls individual sentences, tables, and figure regions, so you can tokenize a phrase, blur a chart section, or replace exact numbers with ranges.
Write policies in natural language, defining purpose / role-based access controls.
Vault
Encryption and structure-preserving tokenization for high-throughput pipelines.
Multiple availability zones with configurable data residency per tenant, isolated to meet compliance needs.
Intercepts prompts, responses, and retrievals to prevent sensitive information from leaking into or out of models. Works with RAG pipelines, copilots, and AI APIs to apply redaction and just-in-time hydration based on policy.
Extends policy enforcement to operational systems like Salesforce, SAP, or Workday. Sensitive fields are tokenized or masked in exports, reports, and syncs, with originals revealed only to authorized roles.
Data Firewall
Follow the journey of sensitive data through our protection system
Sensitive documents enter the system in their original form, containing unprotected personal and financial information
Raw Data
Unfiltered data enters the system
EMPLOYMENT AGREEMENT
Confidential Document
This Employment Agreement ("Agreement") is entered into on January 15, 2024 between TechCorp Industries ("Company") and Sarah Chen ("Employee").
1. Position and Compensation
The Employee shall serve as Senior Software Engineer. The Employee's annual base salary shall be $145,000, payable in accordance with the Company's standard payroll schedule.
In addition to base salary, the Employee is eligible for an annual performance bonus of up to $20,000 based on individual and company performance metrics.
2. Contact Information
Employee Email: sarah.chen@techcorp.com
Employee ID: EMP-2024-1847
3. Benefits
The Employee shall be entitled to participate in all Company benefit programs, including health insurance valued at $12,000/year and 401(k) matching.
4. Confidential Information
Employee acknowledges access to proprietary information including Q4 revenue projections of $2.8M and strategic product roadmaps classified as trade secrets.
Employee Signature
Date: January 15, 2024
HR Director
Date: January 15, 2024
AI automatically identifies sensitive data types like names, salaries, and financial figures based on configurable policies
Detection
Firewall identifies and classifies sensitive information
EMPLOYMENT AGREEMENT
Confidential Document
This Employment Agreement ("Agreement") is entered into on January 15, 2024 between TechCorp Industries ("Company") and Sarah Chen ("Employee").
1. Position and Compensation
The Employee shall serve as Senior Software Engineer. The Employee's annual base salary shall be $145,000Auto-detected by policy, payable in accordance with the Company's standard payroll schedule.
In addition to base salary, the Employee is eligible for an annual performance bonus of up to $20,000 based on individual and company performance metrics.
2. Contact Information
Employee Email: sarah.chen@techcorp.com
Employee ID: EMP-2024-1847
3. Benefits
The Employee shall be entitled to participate in all Company benefit programs, including health insurance valued at $12,000/year and 401(k) matching.
4. Confidential Information
Employee acknowledges access to proprietary information including Q4 revenue projections of $2.8M and strategic product roadmaps classified as trade secrets.
Employee Signature
Date: January 15, 2024
HR Director
Date: January 15, 2024
Detected sensitive data is tokenized, protecting it from unauthorized users. We store the raw values in an encrypted vault
Encryption
Policies applied based on data classification
EMPLOYMENT AGREEMENT
Confidential Document
This Employment Agreement ("Agreement") is entered into on Contract Date A between TechCorp Industries ("Company") and Person A ("Employee").
1. Position and Compensation
The Employee shall serve as Role A. The Employee's annual base salary shall be Salary AEncrypted for compute, payable in accordance with the Company's standard payroll schedule.
In addition to base salary, the Employee is eligible for an annual performance bonus of up to Salary B based on individual and company performance metrics.
2. Contact Information
Employee Email: Email A
Employee ID: Employee ID A
3. Benefits
The Employee shall be entitled to participate in all Company benefit programs, including health insurance valued at Salary C and 401(k) matching.
4. Confidential Information
Employee acknowledges access to proprietary information including Q4 revenue projections of Q4 Metric A and strategic product roadmaps classified as trade secrets.
Employee Signature
Date: Contract Date A
Role B
Date: Contract Date A
Authorized users see hydrated values while third-parties, AI systems, and vendors only see tokenized data
Secure Data
Ready for AI
EMPLOYMENT AGREEMENT
Confidential Document
This Employment Agreement ("Agreement") is entered into on Contract Date A between TechCorp Industries ("Company") and Person A ("Employee").
1. Position and Compensation
The Employee shall serve as Role A. The Employee's annual base salary shall be Salary AAuthorized users see values, payable in accordance with the Company's standard payroll schedule.
In addition to base salary, the Employee is eligible for an annual performance bonus of up to Salary B based on individual and company performance metrics.
2. Contact Information
Employee Email: Email A
Employee ID: Employee ID A
3. Benefits
The Employee shall be entitled to participate in all Company benefit programs, including health insurance valued at Salary C and 401(k) matching.
4. Confidential Information
Employee acknowledges access to proprietary information including Q4 revenue projections of Q4 Metric A and strategic product roadmaps classified as trade secrets.
Employee Signature
Date: Contract Date A
Role B
Date: Contract Date A
The tokenization preserves fundamental structure so LLMs can process the data without ever accessing sensitive information.
Remains AI-Readable
Prevent material non-public information and customer data leaks across research reports, trading communications, and third-party analytics.
Financial Services
Prevent trial outcomes and PHI leaks across clinical workflows, research chats, and CRO exchanges.
Biotech and Healthcare
Prevent classified information and CUI leaks across intelligence reports, contractor exchanges, and AI-assisted workflows.
Government and Defense
Why Velum
| Feature | Velum | DLP | DSPM | LLM Gateway | Tokenization Vault |
|---|---|---|---|---|---|
| Understands unstructured meaning (outcomes, findings, MNPI, privilege) | entities/keywords | inventory only | rules/keywords | structured fields only | |
| Enforces at runtime, per request (in-path) | mostly perimeter/endpoint | at model API only | |||
| Span-level transforms (tokenize • abstract) | coarse block/redact | filter/deny prompts | field tokenization | ||
| Preserves data utility (least-privilege views) | often blocks | may degrade prompts | loses meaning | ||
| Audit-ready evidence (per-decision logs) | alerts | asset logs | request logs | access logs |