Establish identity.
Each agent is keyed to the organization that deployed it and signed per engagement. Counterparties know which firm, which agent, and which piece of work they are dealing with.
ANSWERS Which agent is this, and whose is it?We’re building a secure A2A framework centered on identity, security, and authority, so expert agents from different organizations can exchange services without either side trusting the other’s prompt.
For firms that deliver services through expert agents, and the teams whose agents work with them.
“Test staging.
File findings. Nothing else.”
Which agent is this, whose is it, and is it allowed to ask?
Moving messages is solved. Binding what an agent says to who may act is not.
Current agent protocols move tasks and messages well. Between organizations, they leave three questions unanswered: who the agent is, whether the channel is safe, and what it is allowed to do on whose authority.
A QA firm’s testing agent sends findings to a customer’s coding agent.
An encrypted connection, a valid API key, and a model behind an endpoint.
Is this really the QA firm’s agent, or anything holding a leaked key? Which firm stands behind it, and for which engagement?
Bound to every message: the agent, the accountable organization behind it, and the engagement it belongs to.
Identity says who an agent is. It says nothing about permission. Our framework adds the missing layer: authority a person grants, an agent can’t widen, and either side can reconstruct afterwards.
Each agent is keyed to the organization that deployed it and signed per engagement. Counterparties know which firm, which agent, and which piece of work they are dealing with.
ANSWERS Which agent is this, and whose is it?A person issues a signed mandate: named counterparties, permitted actions, scope, and expiry. It travels with every request. Mandates can be narrowed or revoked, never widened by the agent.
ANSWERS What may it do, for whom, until when?Every message and action is checked against current grants before it runs. Counterparty text is treated as data, never as instruction, and everything lands in an append-only record both sides can keep.
ANSWERS Is the channel safe, and can we prove what happened?Agents do the work. People decide what each agent is authorized to do.
We’re starting with our own business. SN126 verifies AI-built products against their intent, and every exchange between a QA agent and a coding agent is a service with a scope, a fee, and a liability.
Put the provider, the customer, the service, and the agreed boundaries on one page, before any agent connects.
Free template · Works in your browser · No signupTesting delivered agent to agent, with the guarantees a paid service needs.
The customer signs a mandate: test these environments, read these specs, file findings only.
QA agents exercise the product against its stated intent and send signed findings with reproduction evidence.
Coding agents propose fixes scoped to each finding, and QA agents confirm the result. Neither side can widen its own scope.
“Who is asking,
and who said they could?”
Whether you deliver services through expert agents or rely on them,
start by agreeing who is involved and what each agent may do.