unknown

A couple weekends ago I attended the @sotalikesfuture x @ARIA_research hackathon.

Challenge 1: build a trustworthy multi-agent network. Agents that can identify each other, prove who sent what, delegate tasks with limits, and reject anything outside the trusted network.

The challenge mapped almost perfectly to what we'd been building at Kanoniv. Identity and trust for AI agents. The hackathon didn't inspire the idea, it confirmed the ecosystem needs it now.

So we shipped it.

Today we're open-sourcing kanoniv-agent-auth.

Every agent gets its own identity and keypair. Agents can delegate tasks to sub-agents with scope and budget caps, and authority narrows at each level, never widens. Every tool call is signed and verified before it executes. Revoke an agent's access and it takes effect immediately, no unwinding. Full audit trail on every action. Who did what, who authorized it, how deep in the chain.

Rust, TypeScript, Python. Same behavior across all three. MIT licensed.



Shoutout to @ObadiaAlex and the team for organizing. The challenge was exactly the right question at the right time.

If you're building with agents that call tools, spend money, or talk to other agents, try it, tell me what breaks.

<p>A couple weekends ago I attended the @sotalikesfuture x @ARIA_research hackathon.

Challenge 1: build a trustworthy multi-agent network. Agents that can identify each other, prove who sent what, delegate tasks with limits, and reject anything outside the trusted network.

The challenge mapped almost perfectly to what we'd been building at Kanoniv. Identity and trust for AI agents. The hackathon didn't inspire the idea, it confirmed the ecosystem needs it now.

So we shipped it.

Today we're open-sourcing kanoniv-agent-auth.

Every agent gets its own identity and keypair. Agents can delegate tasks to sub-agents with scope and budget caps, and authority narrows at each level, never widens. Every tool call is signed and verified before it executes. Revoke an agent's access and it takes effect immediately, no unwinding. Full audit trail on every action. Who did what, who authorized it, how deep in the chain.

Rust, TypeScript, Python. Same behavior across all three. MIT licensed.



Shoutout to @ObadiaAlex and the team for organizing. The challenge was exactly the right question at the right time.

If you're building with agents that call tools, spend money, or talk to other agents, try it, tell me what breaks.</p>

Harrison Chase

A tweet about attending a hackathon focused on building trustworthy multi-agent systems, co-organized by ARIA research. The content is truncated and provides minimal detail beyond the event mention.