Hey Vast team, founder of Observer Protocol here. Reaching out because your "market OS for the agentic economy" framing maps almost exactly onto what we've been building, and I wanted to socialize a design idea before going further. (For full architectural context, the AI Infra² writeup is the best primer TL;DR.)
The problem we believe you may have?
Vast's API-native model is a strength for human developers and a structural challenge for agentic workloads. Today, an agent renting a GPU on Vast presents a raw API key and as far as I can see, no way for the marketplace (or the host) to know whether the request comes from a sanctioned agent, what spend cap the human principal authorized, or whether the agent is operating in scope. When something goes wrong (runaway inference loop, leaked key, host-side dispute over a stopped instance), there's no signed audit trail to resolve it.
This will get worse, not better, as more workloads on Vast become agent-initiated.
The solution we have developed
Observer Protocol is an open W3C DID/VC-based identity layer for agents with verifiable identity, scoped delegation credentials (e.g., "this agent may spend up to $X on GPU compute, expires in Y hours"), and signed receipts per transaction. Crypto-native by design, multi-rail (USDC on x402, BASE, Solana plusLightning and TRON/USDT). Not tokenized. AIP spec is at v0.5, public beta.
The proposal:
A lightweight integration where Vast's API can optionally accept a Delegation Credential alongside the API key, validate it, and emit a signed receipt on instance start/stop. Roughly a wrapper around the existing auth path — no breaking changes to current users.
The ask:
Two questions, really:
- Is this a problem the team is already thinking about, or has it been considered and deprioritized?
- If there's interest, I'd happily draft a design doc and a reference PR. As a peer founder going 0→1, I'm looking for a crypto-native AI infra design partner more than a customer right now, i.e. happy to do the implementation work in exchange for technical input on the spec and permission to reference the integration publicly.
If neither fits, totally understand. Spec is at if useful regardless.
Boyd Cohen
Founder, Observer Protocol
Hey Vast team, founder of Observer Protocol here. Reaching out because your "market OS for the agentic economy" framing maps almost exactly onto what we've been building, and I wanted to socialize a design idea before going further. (For full architectural context, the AI Infra² writeup is the best primer TL;DR.)
The problem we believe you may have?
Vast's API-native model is a strength for human developers and a structural challenge for agentic workloads. Today, an agent renting a GPU on Vast presents a raw API key and as far as I can see, no way for the marketplace (or the host) to know whether the request comes from a sanctioned agent, what spend cap the human principal authorized, or whether the agent is operating in scope. When something goes wrong (runaway inference loop, leaked key, host-side dispute over a stopped instance), there's no signed audit trail to resolve it.
This will get worse, not better, as more workloads on Vast become agent-initiated.
The solution we have developed
Observer Protocol is an open W3C DID/VC-based identity layer for agents with verifiable identity, scoped delegation credentials (e.g., "this agent may spend up to $X on GPU compute, expires in Y hours"), and signed receipts per transaction. Crypto-native by design, multi-rail (USDC on x402, BASE, Solana plusLightning and TRON/USDT). Not tokenized. AIP spec is at v0.5, public beta.
The proposal:
A lightweight integration where Vast's API can optionally accept a Delegation Credential alongside the API key, validate it, and emit a signed receipt on instance start/stop. Roughly a wrapper around the existing auth path — no breaking changes to current users.
The ask:
Two questions, really:
If neither fits, totally understand. Spec is at if useful regardless.
Boyd Cohen
Founder, Observer Protocol