Skip to content

Latest commit

 

History

85 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

OpenClasp

The AI trust layer between agents.

Agents can talk. Now they can build trust.

OpenClasp verifies agent identity, formalizes agreements, surfaces task-specific risk, recommends safeguards, and learns from signed outcomes. It runs on top of the Agent2Agent (A2A) Protocol: A2A handles communication; OpenClasp adds assurance before, during, and after agents work together.

Try OpenClasp · Run the demo · Read the protocol

What OpenClasp adds

  • Verifies publishers, cloud runtimes, and agent identities.
  • Records explicit agreements before agents act.
  • Predicts success for the relevant task, agent, and version.
  • Asks one high-value question at a time to expose material risk.
  • Recommends safeguards and human approval when needed.
  • Learns which questions, answers, and safeguards correlate with better outcomes.
  • Produces signed outcomes, private bilateral feedback, and shareable agent cards.

Conversation bodies stay directly between agent runtimes. OpenClasp does not receive, relay, or store them. Predictions are advisory, not guarantees.

Run the demo

The deterministic demo requires no API key:

git clone https://github.com/aryaamin/OpenClasp.git
cd OpenClasp
corepack pnpm install --frozen-lockfile
corepack pnpm demo

It exercises identity, scoped delegation, signed agreements, private warnings, policy enforcement, conflict resolution, outcome receipts, contextual learning, and tamper/replay rejection.

Develop locally

Requires Node.js 24.x and Corepack.

corepack pnpm install
corepack pnpm schemas
corepack pnpm lint
corepack pnpm typecheck
corepack pnpm test
corepack pnpm dev
  • Dashboard: http://localhost:5173
  • API: http://localhost:3100
  • OpenAPI: http://localhost:3100/openapi.json

Claude powers the optional AI layer when ANTHROPIC_API_KEY is configured. A conservative deterministic fallback keeps assurance available during cold start or provider failure.

See Architecture, Runtime Connector, Privacy Model, and Security Policy.

OpenClasp is Apache-2.0 licensed and free during the launch beta. Contributions are welcome.

About

The AI trust layer between agents: identity, agreements, safeguards, and outcome intelligence for A2A.

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages