@verifiedstate/openai-agents
v0.2.0
Published
VerifiedState integration for OpenAI Agents — cryptographic receipts for every agent assertion
Maintainers
Readme
@verifiedstate/openai-agents
OpenAI Agents Hand Off Context. Not Proof.
VerifiedState integration for OpenAI Agents. Adds cryptographic verification receipts to every assertion your OpenAI Agents agent makes.
Install
npm install @verifiedstate/openai-agentsQuick Start
import { VerifiedStateOpenaiAgents } from '@verifiedstate/openai-agents';
const memory = new VerifiedStateOpenaiAgents({
apiKey: 'vs_live_...',
namespaceId: 'your-namespace-uuid',
});
// Store and verify a fact from your OpenAI Agents agent
const { artifact, assertions, receipts } = await memory.storeAndVerify(
'Customer prefers dark mode and uses PostgreSQL'
);
console.log(receipts[0].status); // "verified"
console.log(receipts[0].final_confidence); // 0.94
// Query verified memory
const results = await memory.query('What database does the customer use?');
console.log(results.answerable); // true
console.log(results.channels_used); // ["sparse", "dense"]Why
Every fact your OpenAI Agents agent stores should come with proof — a signed receipt showing what the claim is, what evidence supports it, and whether it's been challenged or superseded.
Without verification receipts, you have the agent's output. You don't have the reasoning.
API
store(content, sourceId?)
Store content from your OpenAI Agents agent.
extract(artifactId)
Extract structured assertions.
verify(assertionId)
Verify an assertion, produce a signed receipt.
storeAndVerify(content, sourceId?)
All-in-one: store, extract, verify.
query(queryText, limit?)
Six-channel retrieval with abstention.
health()
Namespace health metrics.
Links
License
MIT
