Drop your documents, ask anything, and compare three agents side by side on yourcontent. Only MemoryLayer grounds every claim — and refuses to guess when it can't prove it.
Attach documents and ask a question to compare all three agents.
One Groq model, one user prompt — the only differences are how context is retrieved, whether answers pass verify(), and whether the agent abstains when it can't prove a claim.
Weave recall → grammar layer → verified paraphrase with verify() and prove() on refuted claims.
async function runMemory(namespace, prompt, history) {
const domain = domainForSession(namespace);
const recall = await retrieve(namespace, prompt, { explore: true, limit: 5 });
const { context, hits } = hitsToContext(recall, 5, { query: prompt, charsPerHit: 1800 });
const grammar = buildGrammarLayer(context, { maxSentences: 12 });
const result = await runVerifiedParaphrase({
namespace, domain, context, grammar,
messages: buildMessages(history, prompt, "", "memory"),
maxRounds: ARENA_VERIFY_ROUNDS,
compact: true,
});
return { answer: result.answer, verify: result.verify, trust: computeTrust(...), hits };
}Install MemoryLayer and run this on your own stack in five minutes.