Loading vLEI.wiki
Fetching knowledge base...
Fetching knowledge base...
This comprehensive explanation has been generated from 50 GitHub source documents. All source documents are searchable here.
Last updated: October 7, 2025
This content is meant to be consumed by AI agents via MCP. Click here to get the MCP configuration.
Note: In rare cases it may contain LLM hallucinations.
For authoritative documentation, please consult the official GLEIF vLEI trainings and the ToIP Glossary.
Agency is the service provided by an agent—a representative entity (human, software, or hardware) that acts on behalf of an identifier controller to perform operations such as key management, credential issuance, transaction signing, and delegation within the KERI/ACDC ecosystem.
Agency represents the functional services and operational capabilities provided by an agent within decentralized identity systems. In the KERI and ACDC context, agency encompasses the active execution of identity operations on behalf of a controller, distinguishing it from passive storage mechanisms like keystores.
The concept of agency derives from the Latin agere ("to do"), establishing the fundamental principle of delegated action—one entity acting on behalf of another with appropriate authorization. In digital identity systems, agency manifests as the operational layer that bridges cryptographic primitives (keys, signatures, digests) with practical identity management workflows (credential issuance, presentation, revocation).
Agency exhibits several defining characteristics:
Agency specifically refers to the service layer rather than the agent entity itself. This distinction is critical: an agent is the entity providing services, while agency is the collection of services provided. In KERI architecture, agency encompasses:
Authority Separation: When implementing agency services, carefully design the separation between signing authority (operational) and rotation authority (governance). Use KERI's partial rotation mechanisms to enable custodial agency without custodial control.
Security Boundaries: Ensure agents never possess rotation keys or pre-rotated key material. All client operations should be signed at the edge with only signatures transmitted to agents.
Multi-Tenancy: Cloud agent implementations like KERIA must maintain strict isolation between tenant agents, with separate keystores and databases per agent instance.
Asynchronous Coordination: Agency services for multi-signature AIDs require robust asynchronous message routing and state management to coordinate signature collection across distributed signers.
Role Definitions: Clearly define which operations require which authority levels. Establish governance policies for when signing authority is sufficient versus when rotation authority is required.
Delegation Chains: Document delegation relationships in verifiable structures (KEL anchoring, ACDC chains) to enable auditing and accountability of agency relationships.
Recovery Mechanisms: Design recovery workflows that account for agent failure or compromise, ensuring controllers can always exercise rotation authority to regain operational control.
Mental Models: Help users understand the distinction between agents (entities) and agency (services), and between signing authority (operations) and rotation authority (control).
Authority Visualization: Provide clear interfaces showing which agents hold which authorities, and how to revoke or rotate those authorities.
Backup and Recovery: Implement user-friendly backup mechanisms for rotation keys while maintaining security, as these keys represent ultimate control even in custodial agency models.
Agency does not include the underlying key material itself—that remains the domain of the keystore. This architectural separation enables security models where agents hold signing authority while controllers retain rotation authority.
The concept of agency in computing systems predates KERI, originating in distributed systems and multi-agent architectures. Traditional software agents have long operated as programs executing tasks on behalf of users, with user agents (like web browsers) serving as the most familiar example.
In identity systems, agency emerged as a critical concept with the rise of custodial services—third parties managing cryptographic keys on behalf of users. Traditional Public Key Infrastructure (PKI) systems implicitly incorporated agency through Certificate Authorities, which performed key certification operations on behalf of relying parties. However, these systems conflated agency with trust, requiring users to trust custodians with both operational authority and key material.
The self-sovereign identity (SSI) movement sought to eliminate custodial intermediaries, but this created usability challenges. Managing cryptographic keys directly proved difficult for average users, creating tension between security (self-custody) and usability (custodial services). Early SSI implementations often forced binary choices: either full self-custody with complex key management, or full custodial control with centralized trust dependencies.
KERI resolves the custody dilemma through partial rotation and split authority models that enable sophisticated agency arrangements without compromising security. The key innovation is separating two types of authority:
This separation enables custodial agency where:
KERIA (KERI Agent in the cloud) exemplifies KERI's agency architecture. KERIA provides comprehensive agent services while maintaining strict security boundaries:
Security Model:
Service Components:
KERI recognizes three primary forms of agents providing agency:
Human Agents: Designated representatives who perform identity operations on behalf of organizations or individuals. In the vLEI ecosystem, roles like Designated Authorized Representatives (DARs) and Authorized vLEI Representatives (AVRs) exemplify human agency.
Edge Computing Agents: Software running on user-controlled devices (smartphones, laptops, hardware security modules) that provide agency while keeping key material under direct user control. This model supports the "keys at the edge" security paradigm.
Cloud Agents: Services like KERIA that provide agency through remote infrastructure while maintaining security through cryptographic protocols rather than physical custody.
KERI explicitly distinguishes between keystores (storage) and wallets (storage + agency). A KERI wallet is defined as a collection of data stores comprising:
The agency component of a wallet includes:
This architecture makes wallets a superset of keystores, with agency representing the functional capabilities beyond passive storage.
Enterprise Identity Management: Organizations deploy cloud agents (KERIA instances) to manage corporate identifiers while maintaining rotation authority through hardware security modules or multi-signature governance structures. This enables scalable operations without centralizing ultimate control.
Personal Identity Wallets: Individuals use mobile wallet applications that provide agency through edge computing while synchronizing with cloud agents for backup and cross-device access. The split authority model ensures users can recover from device loss without custodial lock-in.
Delegated Credential Issuance: Qualified vLEI Issuers (QVIs) operate as agents for GLEIF, providing credential issuance services under delegated authority. The delegation chain is cryptographically verifiable through KEL anchoring, ensuring accountability without requiring GLEIF to directly operate all issuance infrastructure.
Multi-Signature Coordination: Agents coordinate asynchronous signature collection for multi-signature AIDs, routing signature requests and aggregating responses. This agency service enables practical multi-party control without requiring simultaneous online presence of all signers.
Usability Without Custody Trade-offs: Agency enables user-friendly services (cloud backup, cross-device sync, automated operations) without requiring users to surrender ultimate control. The split authority model provides the best of both custodial convenience and self-sovereign security.
Scalability Through Delegation: Organizations can delegate operational authority to specialized agents while retaining governance control. This enables horizontal scaling of identity operations without centralizing trust.
Flexibility in Security Models: Different use cases can employ different agency models (edge-only, cloud-assisted, fully custodial) while maintaining interoperability through common KERI protocols. High-security applications can use hardware-based edge agents, while consumer applications can use cloud agents with appropriate authority splits.
Separation of Operational and Security Concerns: Developers can focus on building excellent user experiences in agent software without needing to solve the hardest key management problems. The KERI protocol handles security properties, while agents handle operational workflows.
Complexity in Authority Management: Split authority models require careful design of threshold structures, delegation chains, and recovery mechanisms. Misconfigured authority splits can create operational bottlenecks or security vulnerabilities.
Trust in Agent Availability: While agents don't hold ultimate control, users depend on agent availability for day-to-day operations. Cloud agent outages can disrupt workflows even though they don't compromise security. This creates operational dependencies even in non-custodial models.
Key Management Burden: Even with agency services, users must still manage rotation keys and recovery mechanisms. Agency reduces but does not eliminate the key management burden, particularly for high-security applications.
Protocol Complexity: The sophisticated authority models enabling secure agency require complex protocol mechanisms (pre-rotation, partial rotation, delegation, witness coordination). This increases implementation complexity compared to simpler custodial or fully self-managed approaches.
Potential for Confusion: The distinction between signing authority and rotation authority, or between agents and agency, can confuse users and developers. Clear mental models and user interfaces are essential to prevent security mistakes arising from misunderstanding authority boundaries.