SERAA Cortex

SERAA Cortex is a control plane for building, orchestrating, registering, governing, and monitoring AI agents across cloud environments. It is the Act layer of Covasant's SERAA platform. The platform is vendor-neutral: it governs agents built in Cortex alongside agents built in Gemini Enterprise, Azure AI Foundry, AWS Bedrock, and Databricks AgentBricks. The primary users are AI developers, solutions architects, governance leads, platform administrators, and AI operations managers.

Lifecycle Modules

• Agent Work Bench is the build environment. It supports no-code and pro-code agent creation side by side, with the pro-code option running on LangGraph. Builders configure persona, base model, system prompt, and an optional retrieval-augmented generation (RAG) knowledge base, then test in an integrated playground.
• Multi-Agent Orchestration (MAO) is a drag-and-drop canvas for wiring agents into multi-step workflows, with dependency management, error recovery, and a different model selectable at each node.
• Agent Registry is the organization-wide catalog of agents, including those built elsewhere. Agents carry semantic versioning with version diffs. They move through Dev, QA, Staging, and Production by a promotion workflow that cannot be skipped. Each promotion requires a checklist-based approval gate with a configurable readiness threshold.
• Agent Control Tower is the production operations view. It attributes cost by agent, model, user, provider, and tenant, enforces budget caps, and logs every large language model (LLM) call with its token count and cost. It also enforces guardrails and includes a kill switch for a misbehaving agent.
• Agent Test Bench (ATB) is in beta and is tracked separately from the four modules. It runs LLM-as-judge evaluation and digital-twin simulation before an agent reaches production, with scenarios drawn from expert, novice, and adversarial personas.

Agent Work Bench, Agent Registry, and Agent Control Tower are also available as standalone offerings. System Config is the administration surface for users, models, and runtimes.

Technical Specifications

• Protocols: Every external tool integration is a Model Context Protocol (MCP) server registered in a central MCP Hub. Agent-to-Agent (A2A) protocol support lets agents call other agents, including agents on third-party platforms, as sub-agents.
• Models: The Model Hub lists more than 100 LLMs from Google, Anthropic, OpenAI, Azure, AWS, and Databricks, according to the vendor.
• Runtimes: Agents deploy to Kubernetes (GKE, AKS, EKS), GCP Agent Engine, Cloud Run, Azure Functions, and Foundry Agent Service. A Plugin SDK extends Cortex to proprietary runtimes, including on-premises and virtual private cloud (VPC) edge.
• Knowledge stores: These include Vertex AI Search, Azure AI Search, Amazon Kendra, OpenSearch, S3 Vector, Mosaic AI Vector Search, and FAISS.
• Access and security: Scope-Based Access Control (SBAC) uses nested System, Organization, and Project scopes. Single sign-on works with Microsoft Entra ID, GCP IAM, AWS Cognito, and SAML. Tenant isolation and an immutable activity audit log are built in. Guardrail policies cover content safety, personally identifiable information (PII) detection and redaction, and General Data Protection Regulation (GDPR) and Health Insurance Portability and Accountability Act (HIPAA) compliance.
• Hosting: The platform runs on Google Cloud, AWS, or Azure.

Name Change

SERAA Cortex was previously named CAMS (Covasant Agent Management Suite). The vendor states that it is the same product and roadmap under a new name, and that existing contracts, APIs, and deployments are unaffected. Documentation, reviews, and listings under the CAMS name describe earlier versions of the product.