QDEX / MrProbe covers MrProbe, an AI agent assurance platform from QDEX Consulting, an AI security and enterprise AI consultancy. MrProbe is for teams putting AI agents into production who need to know, before release, whether an agent can be attacked and whether it actually does its job well.
QDEX also delivers AI red teaming, LLM penetration testing and AI engineering services around the platform.
At A Glance
| Vendor | QDEX Consulting (maker of MrProbe) |
|---|---|
| Category | AI agent red teaming and functional evaluation |
| Deployment | Hosted app at mrprobe.ai, with a self-hosted ATC path in the customer’s cluster |
| Agentic AI | Adversarial testing of AI agents and runtime policy for agent requests |
| Certifications | ISO/IEC 27001:2022, CSA STAR Level 1 |
| Founded | 2019 |
| Market classification | AI Model Security |
What QDEX / MrProbe Does
An AI agent can fail in two quite different ways: someone can trick it into leaking data or taking actions it should not, or it can simply give wrong, ungrounded or biased answers. MrProbe tests for both and puts them in one report. The security pillar runs a catalogue of attack techniques against the agent and gives each finding a verdict; the functional pillar scores the agent against golden datasets the customer owns. Either pillar can block a deployment when a threshold fails, which turns testing into a release gate rather than a one-off audit.
Testing covers the period before release, and a separate control path takes over in production.
Products
MrProbe (LENS)
The testing product, aimed at teams that need evidence an agent is safe and fit to ship.
- More than 648 adversarial techniques across 17 attack categories, including prompt injection, jailbreaking, data exfiltration, system-prompt leakage, excessive agency, RAG exploitation and supply-chain abuse
- Multi-stage adaptive probing that changes the next attempt based on the previous answer
- Findings marked proven, suspected or defended, with evidence attached
- A functional scorecard across 20 quality dimensions, such as coherence, hallucination, groundedness, toxicity, PII handling, context retention and bias
- Repeat runs against a golden set to catch regressions
- Country-gated PII coverage across 16 jurisdictions
MrProbe ATC
For organisations that need control over AI traffic once agents and assistants are live. ATC is the runtime companion to MrProbe’s pre-release testing: it applies policy to each live AI request and decides whether to allow, mask or block it, across chat, agents and developer tools. Its views cover:
- Spend, threats and active agents in one overview
- Security events for blocked, redacted and warned requests
- Cost and budget tracking
Sniffr.ai
Sniffr.ai reviews pull requests on architecture, requirements and security policy, giving developers feedback they can act on. It is the code layer of the QDEX stack, with MrProbe as the agent layer, and teams use both when a repository also ships an agent.
How It Works
Testing follows a configure, select and execute loop that can run as a consulting engagement or as a CI job.
| Component | What it does |
|---|---|
| Hosted app | Teams register agents, start campaigns and read reports at mrprobe.ai. |
| Testing target | MrProbe runs against agents the customer hosts, including in-cluster deployments. Each engagement scopes the endpoints, the data the probe may touch and the frameworks the report maps to, so production data stays where it is. |
| ATC control plane | On the self-hosted path, prompts, logs and the ledger stay in the customer’s cluster (“You run the plane”). This suits teams that cannot route prompts through a vendor’s multi-tenant proxy. |
| CI gate | CI calls the MrProbe API and fails the build when a gate is not met. |
| Framework mapping | Findings map to 30 frameworks across 16 jurisdictions, such as the OWASP LLM Top 10, NIST AI RMF, the EU AI Act, GDPR, HIPAA, PIPL, APPI, PDPA, LGPD and DPDP. |
Agentic AI Capabilities
Agent Attack Testing
Several attack categories target agent behaviour specifically: agent attacks, excessive agency, multi-agent attacks and RAG exploitation. The red teaming service adds RAG poisoning, multi-agent hijacking and output injection, scoped to the categories that fit the customer’s agent and run only against approved endpoints. Engagements deliver an evidence pack, a framework map and remediation notes.
Blue Team Hardening
After a red team engagement, QDEX deploys guardrails, prompt hardening and runtime policy to close the findings, and repeat MrProbe runs against the golden set can then catch any regression.
Integrations
| Area | Integrations |
|---|---|
| Guardrails | Model Armor, Bedrock Guardrails, Azure, NeMo |
| AI platforms | LangChain, LangGraph, Vertex AI, AWS Bedrock, Azure OpenAI |
Company Background
| Founded | 2019. |
|---|---|
| Headquarters | Chennai, India. |
| Other offices | Dubai, United Arab Emirates. |
| Leadership | Gurunathan Gunaseelan (Founder and Head of Innovation) and Lakshmi B (Co-Founder and Head of People). |
| Certifications | ISO/IEC 27001:2022 for the Dubai entity, and CSA STAR Level 1 through a CAIQ self-assessment. |
Note: This overview is compiled from QDEX / MrProbe’s publicly available information as of 3 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.