RAG + Agents

Index Everything. Verify What Matters.

ChatMemoryBuffer that proves what it stored.

The Problem

LlamaIndex retrieves. It doesn't verify what it found. RAG pipelines extract claims from documents and trust them by default. No receipt of what was extracted, no conflict check, no provenance chain.

The Fix

npm install @verifiedstate/llamaindex

Beta: tagging-level integration. Ingestion works via @verifiedstate/sdk with LlamaIndex-specific source tagging. Native LlamaIndex features coming.

typescriptBeta
import { VerifiedChatMemory } from '@verifiedstate/llamaindex';

const memory = new VerifiedChatMemory({
  apiKey: 'vs_live_...',
  namespaceId: 'rag_pipeline',
});

// Use as ChatMemoryBuffer replacement
const engine = new ContextChatEngine({
  chatHistory: memory,
  retriever: myRetriever,
});

What You Get Back

✓ VERIFIED

receipt.json
{
  "receipt_id": "rct_7f3a...",
  "assertion": "User prefers dark mode",
  "status": "verified",
  "confidence": 0.94,
  "signature": "ed25519:a8c3f1...",
  "verified_at": "2026-04-06T14:32:01Z",
  "chain_position": 42
}

Without VerifiedState vs. With VerifiedState

Feature	Without	With VerifiedState
Extracted facts	Trusted by default	Verified with receipts
Source provenance	Lost after retrieval	Full chain maintained
Contradicting docs	Both used equally	Conflicts flagged

FAQ

Does it replace the full LlamaIndex memory?

It replaces ChatMemoryBuffer. Use alongside LlamaIndex retrievers for verified RAG.

Start building with verified memory.

Free: 50,000 assertions/month. No credit card. No trial expiration.

Get Free API Key Read the Docs GitHub →