Authenticated Knowledge & Trust Architecture for A2A Agents , AKTA.

[lukehinds]

Sharing a prototype project I built called "Akta"

https://github.com/RedDotRocket/akta

It's an attempt to enable secure and verifiable auth and delegation between AI agents. It establishes a framework for time-bound capability-based access control, allowing agents to delegate tasks and share resources with fine-grained control. The system leverages concepts from Decentralised Identifiers (DIDs) and Verifiable Credentials (VCs) to create a cryptographically and auditable chain of trust for autonomous agent operations.

The skills and capabilities used to generate the cryptographic verifiable credential are gathered from an Agents A2A card.
In essence, Akta tries to answer what does a "fully autonomous Agent to Agent authorisation grant look like with no humans in the loop"? a.k.a an Agent delegating tasks to another Agent of their own accord. The human presence is derived from their position higher up the chain to their Agents (and the agents they delegate to). There is also a CLI and library for creating keys, vc's, based on A2A AgentCards and their nominated capabilities and skillz!

If you are interested in this idea and want to hack on it with me, let me know. Typical me style, I have way too many uncompleted projects and I am focusing on getting out my main one over the next few weeks. But I do love all this DID stuff and my heart is in this tech, so hopefully this is valuable to someone one out there

FYI, this is prototype, it has no biz being near production and is not audited yet, its a proof of concept that I might develop futher.