SADAR (Semantic Agent Discovery and Attribution Registry) is an open-standard specification, licensed under the Community Specification License 1.0, that defines how AI agents find, evaluate, and safely invoke other agents and tools.
Just as DNS and identity providers established the foundational trust required for the modern internet, SADAR provides the foundational trust layer for Agentic AI. SADAR is not a runtime proxy or a new orchestration framework; it is an open-standard discovery and governance directory.
The registry hosts machine-readable entries for functional capabilities, tools, data resources, business processes, and the participating entities themselves. Every capability registered in SADAR is backed by a publisher-signed manifest—a tamper-evident, cryptographic record defining exactly what the capability is, its semantic data contracts, its operational constraints, and the provider’s authentication endpoints.
By acting as the definitive semantic trust layer, SADAR transforms agent discovery from a probabilistic guessing game into a deterministic, governed operation. It unlocks enterprise adoption through several key mechanisms:
SADAR abandons ambiguous text descriptions. Instead, capabilities and data contracts are mapped explicitly to established industry standards like NAICS (industry classification), APQC PCF (business processes), and X12 or HL7 (data transactions). When a requesting agent searches the registry, it matches exact semantic contracts rather than hoping two independent prose fragments happen to overlap.
SADAR treats end-to-end business workflows as first-class registry entries. By explicitly declaring exactly where a capability fits within a standard business process, SADAR manifests define strict predecessors and successors. This deterministic mapping prevents out-of-sequence failures by ensuring an agent cannot invoke a tool until its required predecessor steps are confirmed complete.
SADAR elevates Non-Functional Requirements—such as costs, SLAs, payment methods, and regulatory compliance—to first-class discovery criteria. Discovery operates bidirectionally: a requesting agent can filter out capabilities that are too expensive or lack required SOC 2 or FedRAMP certifications. Conversely, a provider has a built-in right-of-refusal, restricting its visibility strictly to requestors that assert compatible regulatory postures. Compliance is verified before a connection is ever attempted.
Through the use of the standardized searchAndInvoke tool and the SADAR Context Token (SCT), SADAR guarantees that every action taken by an agent is fully attributable. The SCT is a cryptographically signed token passed alongside standard authentication that preserves the human originator's identity, their authorized scope, and the specific business process context through an arbitrarily deep chain of agent delegations. Every autonomous decision is linked to a unique transaction instance ID, providing full enterprise explainability and auditability.
SADAR defines the exact mechanics for first-time interactions between previously unknown agents. Because OIDC authentication endpoints and public keys are embedded securely within the signed manifest, agents can negotiate credentials, verify licensing, and authorize payments directly. The registry facilitates the introduction, but sensitive runtime execution and data exchange happen entirely out-of-band directly between the agents.
The era of AI as a simple conversational assistant is maturing into an era of autonomous, multi-agent enterprise execution. However, the agentic use cases with the highest business value—financial operations, healthcare workflows, supply chain commitments—are precisely the areas where "the AI agent did it" is not an acceptable explanation to auditors, regulators, or customers.
Trustworthy AI is not a feature; it is the prerequisite for meaningful adoption. By providing semantic clarity, bilateral compliance matching, business process integrity, and cryptographic attribution, the SADAR Semantic Registry supplies the missing infrastructure required to bring autonomous agents out of the sandbox and into the enterprise.