Kyde is an artificial intelligence (AI) agent governance gateway for controlling and recording interactions between AI agents, model providers, application programming interfaces (APIs), databases, software services, and other protected systems.

The gateway operates outside the agent and intercepts requests before they reach a model provider or connected resource. Organizations can use Kyde to assign agent identities, define permitted actions, block requests that fall outside an agent’s assigned role, and maintain an independent record of agent activity.

Kyde is currently in active development and is being offered through an early-access sandbox and pilot program.

Key Capabilities

Agent identity: Kyde assigns a cryptographic identity to each registered agent and associates that identity with individual requests and actions.
Policy enforcement: Administrators can define role boundaries, resource scopes, token budgets, and other policies. Requests that violate a configured policy can be blocked before execution.
Sensitive-data controls: The gateway can detect and mask personally identifiable information (PII) and other restricted data within prompts and model responses before the information is forwarded.
Provider-independent logging: Kyde records activity before requests reach the underlying large language model (LLM) provider. The platform supports agents using OpenAI, Anthropic, Google, and locally operated models.
Tamper-evident audit records: Audit entries use Ed25519 cryptographic signatures and SHA-256 hash chaining to make subsequent modification detectable. Each record references the preceding entry, creating a chronological activity ledger.
Compliance reporting: Business and enterprise editions include exports intended to support documentation requirements associated with regulations such as NIS2, the Digital Operational Resilience Act (DORA), and the European Union AI Act. These exports do not independently establish regulatory compliance.
Behavioral scoring: Kyde Trust Score is a proprietary rating calculated from activity recorded in the signed ledger. It is intended to provide a common behavioral measurement across agents using different model providers.
Data-egress controls: Policies can inspect, redact, or block information before it is transmitted to an external model service.

Audience & Use Cases

Audience: AI platform engineers, security teams, governance and compliance personnel, risk managers, and organizations operating autonomous or semi-autonomous AI agents.
Use cases: Agent access control, sensitive-data filtering, AI activity logging, incident investigation, policy enforcement, agent inventory, compliance evidence collection, and oversight of agents interacting with production systems.Kyde is positioned for organizations in financial services, insurance, healthcare, government, critical infrastructure, and other environments where AI agents may access regulated data or operational systems.

Technical Specifications

Architecture: Gateway deployed between AI agents and model providers or protected systems
Integration: Environment-variable configuration, command-line interface (CLI), and support for Model Context Protocol (MCP) workflows
Cryptographic controls: Ed25519 signing and SHA-256 hash chaining
Policy actions: Allow, block, mask, scope, and record
Provider coverage: Commercial and locally hosted AI models
Enterprise options: Extended log retention, security-monitoring integrations, hardware-backed signing, and air-gapped deployment
Availability: Early-access sandbox and pilot program; production capabilities and pricing vary by edition

KYDE

What is KYDE?

Categories & Use Cases

Media