Skip to main content

July 21, 2026

Nuggets is the trust layer for autonomous AI. We govern AI at the point of execution.
Nuggets ACP Dashboard
Our companion piece, What an Authority Control Plane Is, lays out the category: authority is not access, and proof is where the differentiation sits. This piece is about who actually delivers it. Our Authority Control Plane is built on open standards, not a proprietary registry. Every link in that chain, from the verified person and business to the agent itself, is carried on W3C decentralised identifiers and verifiable credentials, and the authority delegated to an agent travels with it as its own verifiable credential, so it works the same way across clouds, vendors and organisations. Vendor-agnostic by design, not tied to any one platform. Nuggets enforces that authority at the point of execution. Today that runs in production over LangChain. Every authenticated action is checked against the authority the agent was granted, and produces a signed action receipt a partner or a regulator can verify independently, against Nuggets’ published public key, without touching Nuggets’ systems and without trusting Nuggets’ logs. The receipt carries hashes of the sensitive action content rather than the raw data, and is signed so anyone can verify it’s authentic. Whether an agent’s authority is still valid right now is enforced live, at the point the action is attempted, not read off the receipt after the fact. That’s the chain of accountability the category calls for: a real, verified business and a named person standing behind the agent, not a service account with a name attached to it. The receipt proves it. It’s the difference between an agent estate that’s merely seen and one that’s proven.
Nuggets ACP Multi Screen
Authority in Nuggets is granted on a tight scope: which actions an agent may take, spend caps, which targets it can reach, and when the grant expires. High-value actions escalate to a human approver, with timeout and quorum rules. An estate-wide kill switch fails closed — revoke an agent’s authority, or freeze it entirely, and its very next action is denied at the point of execution before anything runs, the denial signed and recorded like any other decision. A policy change becomes an enforceable constraint the moment it’s made, not a ticket in a queue. Blocking an agent’s action at runtime is now common. But agents act across organisations, and the partner or regulator on the other side can’t see inside your systems. Nuggets makes the authority and the proof travel with the agent, so anyone it transacts with can verify it for themselves. langchain-nuggets is the open-source evidence you can run yourself today. The cross-org demo puts one agent to work under authority delegated across two organisations: one ALLOW path, where the action sits inside its granted scope, and four DENY paths where it doesn’t — an out-of-scope tool, an exhausted spend cap, an expired grant, and a revoked grant — with the agent’s intent bound to each request throughout. Revoke the agent’s authority mid-session, and its next action is blocked before execution. Every allowed decision produces a receipt either side can verify independently, without trusting the other’s logs. Nuggets extends the infrastructure a company already runs rather than replacing it. It sits above existing IAM and cloud, binding the identities they manage to the authority, intent, and proof an agent needs before it acts. It’s ISO 27001 certified, built for teams that answer to a regulator as well as a board. No authority, no action. Every decision proven. See it running in the langchain-nuggets repo. If you’re deploying agents into production and need to prove what they’re allowed to do, talk to us.