singulance-amr

Node binding for mneme — a memory filesystem for AI agents. Per-tenant mmap'd vector + temporal + graph store in the .amr format. Drop-in for Qdrant's vector layer. 13× faster than a REST vector DB at equal recall, 7.5× smaller storage, zero servers.

const { MnemeVectorStore } = require('singulance-amr'); // drop-in for QdrantVectorStore
const store = new MnemeVectorStore({ dataRoot: '~/.mneme/data', dim: 1024 });
await store.upsert('org_acme', [{ id: 'm1', vector: emb, payload: { kind: 'pref' } }]);
const hits = await store.search('org_acme', queryEmb, 5); // [{ id, score, payload }]

CLI:

npx amr ingest <dir> --org acme
npx amr recall "your question" --org acme

Full docs, design thesis, benchmarks, and the .amr format spec: https://github.com/amar3012005/mneme

Apache-2.0.