VisIQ Labs builds data loss prevention and runtime governance for enterprise AI. Its platform stops sensitive data leaking into browser AI assistants and checks what AI agents are about to do, which tools they call and which documents they read, before each action runs.
The problem it addresses is that employees paste data into AI assistants and agents act with real credentials, often outside the view of security teams. “Let AI Agents Act. Keep the Enterprise in Control.”
At A Glance
| Vendor | VisIQ Labs (VISIQ LABS LLC) |
|---|---|
| Category | Enterprise AI DLP and runtime governance for AI agents |
| Identities covered | AI agents, sub-agents, agent credentials, MCP servers and tools |
| Deployment | Managed SaaS, or on-premises, sovereign or self-hosted endpoints; browser extension; in-process SDK |
| Agentic AI | Pre-execution policy on agent actions, red teaming, posture scoring |
What VisIQ Does
VisIQ sits between agents and the tools and data they reach, and between employees and the AI assistants they type into. The approach is to find every agent and AI tool, give each agent only the credentials and data it needs, authorise each action before it runs and keep evidence of every decision, including when agents hand tasks to one another. Each decision produces a signed evidence receipt that can be exported for audits under SOC 2, the EU AI Act and DORA, so governance doubles as compliance evidence.
Products
AI DLP And Context Firewall
AI DLP protects staff use of browser AI assistants. Its Context Firewall extension identifies protected values and masks them before a prompt is submitted, with detection running locally in the extension. Adapters cover ChatGPT, Gemini, Microsoft Copilot, Claude and Perplexity.
Runtime Governance
Runtime governance is for teams putting agents into production. It applies controls at the moment an agent acts:
| Action Governance | Checks each tool call against rules before it runs |
|---|---|
| Retrieval Governance | Allows, masks or denies each document on the agent’s read path |
| Approvals | Routes sensitive operations to a person over Slack, Microsoft Teams, email or a dashboard queue |
| Audit Trail | A signed record of each decision |
Agent Red Teaming
Agent Red Teaming tests an agent before attackers do. It copies the deployed agent’s model, system prompt and tool schemas into a sealed sandbox seeded with synthetic canary data, so no real customer records enter the test. New skills are tested first, optionally failing closed, and attacks are organised by MITRE ATLAS with coverage reported per technique.
AISPM
AISPM works out what an agent could reach if it misbehaved. It inventories credentials, tool grants and data store access, ranks the findings so “the widest gaps get closed first”, and maps results to NIST AI RMF, the EU AI Act or ISO 42001.
How It Works
| Component | What it does |
|---|---|
| SDK harness | A single visiq() call wraps an agent executor, and governance wraps each tool’s dispatch method so that a denial stops execution. SDKs ship as @visiq/harness on npm and visiq on PyPI, with a REST API for other languages. |
| Local evaluation | Decisions are made in-process against a cached rule bundle, so no network round-trip is needed except while waiting for a human approval. |
| Monitor then enforce | New agents start in monitor mode, which audits every decision without blocking; enforcement is switched on per agent or per operation from the dashboard. |
| Hosting | The endpoint defaults to the managed SaaS host, and VISIQ_ENDPOINT points it at an on-premises, sovereign or self-hosted deployment instead. |
| Evidence receipts | Each record is hashed with SHA-256, signed with Ed25519, batched into a Merkle tree, timestamped by an RFC 3161 authority (DigiCert) and chained, with the signing key sealed in AWS KMS. |
Identities And Systems Covered
| AI agents | Each agent has an ID, set explicitly or taken from the package name, and is provisioned automatically on first contact |
|---|---|
| Agent credentials | Compartmentalised per agent identity, with data access scoped to match |
| Sub-agents | Cannot exceed the permissions of the principal that invoked them |
| AI inventory | Custom agents, frameworks, MCP servers and tools, including shadow AI |
| Browser AI assistants | ChatGPT, Gemini, Microsoft Copilot, Claude and Perplexity |
Agentic AI Capabilities
| Generated policy | An AI rule engine produces baseline policy automatically, which ALLOW then enforces before each action. |
|---|---|
| Function and trust tier | A person tags each agent with a business function and a trust tier, and baseline rules limit sensitive data categories to the functions that need them. |
| Reviewed proposals | Red teaming verdicts become proposals, such as blacklisting a skill for one agent, that take effect only after a person approves them. |
Integrations
| Area | Integrations |
|---|---|
| Coding agents | Claude Code, Cursor, OpenClaw, Codex |
| Agent frameworks | LangChain, LangGraph, OpenAI Agents SDK, LlamaIndex, Vercel AI SDK, Mastra, VoltAgent, Semantic Kernel (Python and npm) |
| Discovery sensors | Microsoft Intune, CrowdStrike, SentinelOne |
| Data outputs | Datadog, Elastic, Rapid7, Fluency |
| Review channels | Slack, Microsoft Teams |
Company Background
| Legal entity | VISIQ LABS LLC operates the VisIQ platform. |
|---|---|
| Partners | AWS, OpenAI, Microsoft, SentinelOne and Google Cloud. |
| Services | Lab Services helps customers design, build and deploy agentic workflows. |
Note: This overview is compiled from VisIQ Labs’s publicly available information as of 4 November 2026. It is provided for information only. NHI Mgmt Group does not endorse any vendor, and nothing on this page represents the opinions of NHI Mgmt Group.