Built a small open stack so agents on Vercel don’t treat every domain’s llms.txt as gospel.

Problem: tool-calling agents fetch unsigned instructions from domains they’ve never checked.

What shipped (MIT):

• @trustflow/sdk — verify / sign W3C did:web (Ed25519 EdDSA or P-256 ES256; rejects alg: none and HS*)
• @trustflow/cli — npx trustflow init
• @trustflow/vercel-ai-middleware — refuse to parse unsigned llms.txt in the Vercel AI SDK path
• @trustflow/next-plugin — warns in dev when public/llms.txt or public/.well-known/did.json is missing
• Hosted registry: https://trustflow.systems (badge: Verified by AgenticTrust | trustflow.systems)

Try it:

1
npx trustflow init
2
# or
3
pnpm add @trustflow/sdk @trustflow/vercel-ai-middleware

Repo: https://github.com/etienne-source/agent-trust-sdk Verify example: https://trustflow.systems/verify/trustflow.systems

Maker disclosure: I built this. Happy to answer signing / middleware wiring questions.