QS Cortex

QS PLATFORM 10 · QS CORTEX

Available now / Now includes QS SecureGPT

Build agents fast.
Run them safely.
Trace every step.

QS Cortex is the agent runtime of QS Platform 10. Build agents with proven patterns like ReAct, give them extensive tools, and run them where every call is governed and traceable.

Built on the guardrails, observability and multi-model gateway proven in production as QS SecureGPT, and extended to agentic AI with deep access control and governed documents and assets.

HITRUST r2 certified. In production in regulated environments since mid-2023.
AGENTIC AI · ENTERPRISE GOVERNANCE10
QSCortex

Agent runtime

ReAct · planning · multi-agent

Tool access

Entitled tools via MCP

Model gateway

Guardrails · routing

Secure sandbox · Certified packages · Deep access control · Full trace
QS DevX
Tools and traces
QS Sentry
Governed APIs
Private + public
Approved models
Consistent, repeatable, stable and secure agents.
BUILD FAST.
RUN UNDER CONTROL.
Proven agent patterns

ReAct, planning, reflection, tool use and multi-agent

Controlled tool use

Generated tools, discovered by entitlement through MCP

Traceable by design

Every model call and tool call, visible in QS DevX

01

THE ENTERPRISE AGENT

Anyone can build an agent.
Enterprises need agents they can trust.

An agent is a program whose model output controls the workflow. That power is the point, and the risk. QS Cortex makes agents fast to build and safe to run, with the same controls behind every one.

Consistent and repeatable.

Every agent is assembled from the same proven patterns, the same cataloged tools and the same runtime, so the tenth agent behaves like the first.

Stable in production.

Generated tools return structured results and errors the model can act on, so agents reach answers in fewer iterations and fail predictably when they can’t.

Secure by default.

A sandboxed runtime, certified packages only, deep access control over tools, data and documents, and a trace of every step.

THE RIGHT AMOUNT OF AGENCY

Use rules where rules work.
Give agents agency where logic is too complex to program, and govern it the same way everywhere.

02

AGENT PATTERNS

Proven patterns.
Ready to assemble.

Build agents from the patterns that define the state of the art, without wiring each one by hand. Combine them, version them and reuse them across teams.

01 / TOOL USE

Tool use

Select the right API, document search or retrieval tool for each request.

02 / REACT

ReAct

Reason, act, observe and repeat until the answer is grounded in real data.

03 / PLANNING

Planning

Decompose a complex task into steps, then execute them in order.

04 / REFLECTION

Reflection

Score its own output and iterate until it meets a confidence threshold.

05 / MULTI-AGENT

Multi-agent

Coordinate specialist agents, such as eligibility, benefits and claims, under one manager.

01RulesNo model in the loop02Model routerRoutes safe and unsafe traffic03Tool selectionModel picks the right tools04ReAct loopModel controls the iterations05Multi-agentAgents coordinate agentsLess agencyMore agency
Choose the least agency that does the job. QS Cortex governs every level the same way.
03

ARCHITECTURE

Agentic AI on a foundation
of enterprise governance.

QS Cortex brings together a secure agent runtime, governed tool access, documents and assets, and the model gateway proven as QS SecureGPT. QS DevX supplies the tools and keeps the traces; QS Sentry carries every API call.

CONSUMERSBusiness applicationsQS solutions: Contracts · Benefits · AxiomChat and query interfaceAgent API · authenticatedQS CortexAgent runtime and model gateway of QS Platform 10Secure agent runtimeSandboxed execution · certified, vetted packages onlyReActPlanningReflectionTool useMulti-agentAgent managerShort and long-term memoryVersioned agentsConsistent, repeatable, stable and secure by designMCP clientDiscovers only the toolsits identity is entitled toGenerated tool executorValidated inputs · auth handledStructured errors · shaped responsesDocuments and assetsGoverned content life cycle for agentsIngestionEntity extractionMulti-vector searchAgentic RAGDocument access controlModel gatewayevolved from QS SecureGPTEvery model call passes the same controlsGuardrailsModerationObfuscationMulti-LLM multiplexingPolicy-based routingDeep access control · Policy · ObservabilityEvery model call, tool call and guardrail decision is tracedQS DevXDesign and governTool and agent catalogGenerated from API specsEntitlementsWho can use which toolsTraceabilityEvery run, step by stepMCPtracestool callsQS SentryGoverned APIs to enterprise systems and dataPrivate LLMsOn your infrastructurePublic LLMsApproved providers
Conceptual architecture. Deployment configuration determines model access, entitlements and data boundaries.
  1. 01

    Discover

    The agent’s MCP client asks QS DevX for tools and receives only those its identity is entitled to use.

  2. 02

    Reason and act

    The runtime applies the agent’s pattern, calling models through the gateway and tools through QS Sentry.

  3. 03

    Trace

    Every model call, tool call and guardrail decision is recorded and viewable in QS DevX.

04

CONTROLLED BY DESIGN

Extensive tools.
Strictly controlled.

Agents get access to a broad catalog of enterprise tools, but never on their own terms. What they can run, call and see is decided by the platform.

SECURE RUNTIME

A sandbox, not
an open shell.

Agents run in an isolated runtime that allows only libraries and packages certified and vetted by Quantum Gears. Nothing arbitrary is installed or executed.

No unvetted code in production.

GENERATED TOOLS

Tools built from
the API spec.

QS DevX generates each API tool with input validation, authentication, structured errors and responses shaped for the model’s context window.

Fewer iterations to the answer.

ENTITLED DISCOVERY

Agents see only
what they’re allowed to.

The QS Cortex MCP client discovers tools in QS DevX based on permissions. Deep access control extends to documents and assets.

Least privilege for every agent.
Hand-wired toolCall APIOpaque errorGuess and retryOversized responseAnswerQS DevX-generated toolValidated callStructured resultAnswer
Illustrative. Structured results and errors mean fewer wrong turns, fewer model calls and more predictable agents.

05 · TRACEABILITY

Every step traced.
From prompt to API and back.

Each agent run is recorded end to end: the guardrail decisions, the plan, every model call, every tool call through QS Sentry, every document retrieved and the final answer.

Traces are viewable in QS DevX, so developers can debug an agent, auditors can see why it answered as it did, and platform teams can track usage by agent, tool and team.

Explore QS DevX
Agent run · Benefits questionIllustrative traceInput guardrailsgatewayPlanPrivate LLMget_memberTool · QS Sentryget_benefitsTool · QS SentrySearch plan documentsDocumentsReasonPrivate LLMReflect and scorePrivate LLMOutput guardrailsgatewayUser · identity · model · tokens · inputs and outputs recorded for every step
06

FROM QS SECUREGPT

QS SecureGPT is now part of QS Cortex.
QS Cortex goes much further.

QS SecureGPT proved that enterprises can use many models safely. Its multi-model gateway, guardrails and observability now form the governed core of QS Cortex, and QS Cortex builds agentic AI on top of it.

THE MODEL GATEWAY · FROM QS SECUREGPT

Governed access to every model.

  • Multi-LLM gateway and multiplexing across private and public models
  • Guardrails and moderation on inputs and outputs
  • Obfuscation of sensitive data
  • Observability of every model call
  • Role-based access to models and use cases

WHAT QS CORTEX ADDS

Agentic AI as the core competence.

  • Secure agent runtime with ReAct, planning, reflection and multi-agent patterns
  • MCP client that discovers entitled tools in QS DevX
  • Generated API tools with structured errors
  • Governed documents and assets for agents
  • Deep access control and a full trace of every step

Running QS SecureGPT today?
Your models, guardrails and policies carry forward into QS Cortex, and your teams can start building agents on them.

Plan your move to QS Cortex

THE DETAILS

Questions,
answered.

Agent patterns, controls, traceability and the move from QS SecureGPT.

Talk to our team
What is QS Cortex?

QS Cortex is the agent runtime and model gateway of QS Platform 10. It builds and runs AI agents with proven patterns, governed tool access, documents and assets, and full traceability.

What happened to QS SecureGPT?

QS SecureGPT is now part of QS Cortex. Its multi-model gateway, guardrails, moderation, obfuscation and observability form the model gateway inside QS Cortex.

Which agent patterns does QS Cortex support?

Tool use, ReAct, planning, reflection and multi-agent coordination, which can be combined and reused across agents.

How do agents find the tools they can use?

The QS Cortex MCP client discovers tools in QS DevX and receives only those the agent’s identity is entitled to use.

Can agents run arbitrary code or packages?

No. Agents run in a sandboxed runtime that allows only libraries and packages certified and vetted by Quantum Gears.

Why are API tools generated instead of written by hand?

Generated tools handle validation, authentication, errors and response size consistently, so agents take fewer iterations to reach an answer and behave predictably.

Where can we see what an agent did?

Every model call, tool call and guardrail decision is traced and viewable in QS DevX.

Which models can agents use?

Approved private and public models. The model gateway routes each request according to your policies.

Is QS Cortex sold separately?

QS Cortex is included in every QS Platform 10 edition, alongside QS Sentry and QS DevX.

QS CORTEX · PART OF QS PLATFORM 10

Build agents fast.
Run them under control.

See how QS Cortex builds, runs and traces an enterprise agent on your own APIs and documents.