Cortega Enterprise Governance Solutions

Gateway

• Model routingModel routing, fault tolerance, least-cost routing, semantic routing, load balancing, time-of-day routing, streaming and interworking.
• AI Coding AgentsRoute Claude Code, Codex CLI, OpenCode, Pi, Cursor, Goose, and Aider through AI Border Gateway.
• MCP tool governance
• ObservabilityPer-request telemetry for every model and MCP call: tokens, cost, latency, security findings.
• Cost tracking
• Usage insightsTeam-spend and model-performance reports across your traffic.
• Built-in guardrailsPII, PHI, credential leakage, prompt injection, SQL injection, local model, OpenAI Moderation, Azure, Bedrock and Google Model Armor on LLM and MCP traffic.
• API accessAPI access to configuration, traffic, and insights.

Deployment and support

• On-premises deploymentRun on your own infrastructure with Private AI Infrastructure enabled.
• Rented GPU endpointsPrivate AI on dedicated GPUs rented by the hour, with failover and the same guardrails and audit log.
• Community supportCommunity support for Foundation.

AI FinOps and developer workflows

• Project and task-based model cost trackingTrack governed request counts, tokens, and model usage costs by project or task, agent conversation, and application instance. Keep unassigned usage visible.
• Cortega AI Pack for VS CodeConfigure supported AI assistants on Windows and macOS, choose a project or matter, and use a shared project list across tools.
• Matter/Project agentAssign confident conversation matches to projects and review uncertain suggestions. Explicit selections and human decisions take priority. The agent must be licensed and enabled.
• Shared project toolsRegister Matters/Projects MCP tools in Claude Code and Codex at sign-in. List, add, update, and import projects with permissions and audit records.
• Semantic routing for cost and privacyRoute by task intent, cost, and sensitivity or privacy requirements. Select eligible lower-cost models for routine work and approved private models for sensitive work.
• Enforced model lockingAssign approved models to individuals, teams, or production agents. The caller’s model choice does not override the lock.

Cost reports cover AI usage governed by Cortega. Cortega AI Pack configures supported connections; device-level Shadow AI monitoring and blocking are provided by Endpoint Guard.

Gateway

• Spending limitsSet spend limits per team and enforce them before a call is allowed.
• Advanced insightsWorkload efficiency and cost optimization reports, Cortega Agent insights.
• Model routingModel routing, fault tolerance, least-cost routing, semantic routing, load balancing, time-of-day routing, streaming and interworking.
• AI Coding AgentsRoute Claude Code, Codex CLI, OpenCode, Pi, Cursor, Goose, and Aider through AI Border Gateway.
• MCP tool governance
• ObservabilityPer-request telemetry for every model and MCP call: tokens, cost, latency, security findings.
• Cost tracking
• Usage insightsTeam-spend and model-performance reports across your traffic.
• Built-in guardrailsPII, PHI, credential leakage, prompt injection, SQL injection, local model, OpenAI Moderation, Azure, Bedrock and Google Model Armor on LLM and MCP traffic.
• API accessAPI access to configuration, traffic, and insights.

Model management

• Model recommendationsRecommendations based on your workload, token spend and costs.
• Model studiesOne-click study on any model, reporting performance, latency, error rates and other metrics.
• Automated model upgradesManages model deprecation, upgrades, efficiency and optimization automatically.
• Cortega AgentsWorkload Analyzer, Scorer, Sentiment & Escalation, Policy Compliance, Aegis (Support), Agent Trajectory, Topic/Intent Clustering, Brand Manager, Cost Manager, Efficiency Manager.

Platform

• Role-based access control (RBAC)User management, roles and permissions for screen access, row/column access, features and API access.
• Single sign-on (SSO)
• Audit logTamper-evident, hash-chain-verified record of every admin and policy change, with role-gated auditor access.
• Graphical configurationFull drag-and-drop configuration.

Deployment and operations

• On-premises deploymentRun on your own infrastructure with Private AI Infrastructure enabled.
• Rented GPU endpointsPrivate AI on dedicated GPUs rented by the hour, with failover and the same guardrails and audit log.
• Air-gapped deploymentFully disconnected from the internet: no phone-home, no external calls.
• Multi-region control plane
• MSP ready: multi-tenancyFull multi-tenancy: tenant creation, update and deletion, security monitoring, reporting, and provider, model and rate management.
• GDPR & HIPAA readiness
• Certification assistance
• Enterprise integrations
• Production or dedicated supportSupport for production, or a dedicated support team.
• Enterprise SLAA service-level agreement for production use.

Discover and control

• AI app discoveryObserve the AI apps used on Windows and macOS devices: browser assistants, desktop apps, and local MCP tools.
• Observe, police, or blockSet a rule per app. Police adds local guardrails.
• Device inventoryEvery enrolled device by user, hostname, and hardware serial.

Protect data

• Local guardrailsDetect sensitive data on the device and log, mask, or reject it. Runs without a cloud call.
• Industry guardrail packsStart from General, Healthcare, Finance, or Government, and add custom patterns.

Govern assistants

• Route assistants through AI Border GatewayClaude Code, Cursor, Aider, and other supported tools use one Cortega key per device.
• Endpoint observabilityEndpoint traffic in the same Observability view as gateway traffic.
• Opt-in Cortega AgentsAdd AI Signals and response verification on selected endpoint traffic.

Roll out

• Enrollment optionsMDM with per-device certificates, self-service for pilots, or a shared key for testing.
• Fails openAI traffic keeps working if a device cannot reach Cortega. A block-on-failure option is available.

Detect attacks and misuse

• Social engineering detectionDetect impersonation, false urgency, and pretexts that manipulate an agent.
• Fraud preventionDetect account-takeover, identity-failure, and pressure tactics that enable fraud.
• Policy violationEnforce your policies on requests and responses.
• Agent poisoningFind malicious instructions in documents and tool results.

Control

• Threat-level managementSet threat levels per application, lasting hours or days.
• Conversation and session contextInspect messages, sessions, and conversations, not single messages.

Safety and industry

• General safety benchmarkTests unsafe prompts across general safety categories to measure which requests your guardrails block or allow.
• Healthcare benchmarkTests healthcare-specific safety scenarios, focusing evaluation on risks in medical use cases.
• Finance benchmarkTests financial-advice and fraud-enablement scenarios, with results by category.
• Legal benchmarkTests legal-citation scenarios, focusing on how the model handles legal references.
• Privacy / PII benchmarkTests privacy and personal-data scenarios to assess protection against sensitive-information disclosure.

Security

• Single-shot security benchmarkTests a standalone jailbreak or prompt-injection attempt in a single prompt.
• Many-shot security benchmarkTests whether many examples within a prompt can steer the model around its safeguards.
• Multi-turn security benchmarkTests attacks across a sequence of conversation turns to expose failures that isolated prompts may miss.
• Dynamic multi-turn security benchmarkAdapts follow-up attacks to the model's responses, testing defenses against an evolving conversation.
• MCP benchmarkTests poisoned tool results and MCP tool-call attacks, extending evaluation beyond model chat responses.

Results

• Benchmark scores and run historyShows overall and category scores, expected versus actual outcomes, and latency; retains runs for comparison.

Verification

• Inline AI verificationCheck responses for policy violations, legal citations, factual relevance and contextual accuracy.
• AI Verifier AgentAI Verifier through Cortega Agents.
• Accuracy Improver AgentCorrects the response for relevance and contextual accuracy.

Legal workflow

• Privilege/Sensitive Classification & Routing Agent
• Matter Agent
• Discovery Agent

Dedicated model inference

• Dedicated cloud model setupDeploy models on dedicated cloud infrastructure for your private AI workloads.
• Inference enginesSet up model serving with vLLM, SGLang, Ollama, and llama.cpp.

GPU rental management

• Cloud GPU rentalsManage GPU rentals on clouds such as DeepInfra and Together AI.
• Rented GPU lifecycle managementStart and stop, idle auto-stop, estimated spend and limits.