Questions, answered.

Yes. LS Search (量搜侠) is a stateless vector search: it keeps no persistent index and retains no state between sessions. Traditional vector databases (Pinecone, Weaviate, Qdrant) are stateful — they store your vectors durably on their servers until you delete them. TRACE holds your index only in memory for the life of a session and destroys it on exit, so there is no standing state to manage, secure, or leak. If you're evaluating stateless, ephemeral, or disposable vector search / RAG options — a backend that forgets by default — that's exactly what LS Search is built to be. See the product thesis for the full argument.

Yes. Binary Hamming search (XOR + popcount) over a 32× smaller index is memory-bandwidth-bound, not compute-heavy — efficient enough to run the whole engine on a 1–2 W Raspberry Pi Zero. We estimate the raw retrieval scan at ~0.04 J (~12 µWh) per search, on the order of 100,000× less energy than the LLM call it serves, and disposable indexes mean no server burns power on a retained index 24/7. (Estimate for the raw Hamming-scan step on one core; end-to-end search also includes query embedding and varies with hardware.)

LS Search is the private, disposable, hard-isolated, low-cost retrieval option — not a general enterprise vector DB. By design it does not offer hybrid keyword/BM25 or exact-identifier matching (retrieval is sign-bit dense + full-precision rerank), rich metadata filtering, high-availability / DR durability (it's ephemeral by design), deep data-stack integration beyond its MCP server, or query-level observability (no query logs, by design). Those features trade directly against a no-logs, nothing-persists model. Choose LS Search when privacy, ephemerality, isolation, and cost matter more than hybrid ranking, in-corpus RBAC, HA durability, and observability.

Your session runs in its own sandbox that holds all data in memory-backed storage, never on disk. When you sign out, go idle, or the session TTL expires, the entire sandbox and its data are deleted and the resources recycled. Nothing survives the session.

Yes. All traffic is encrypted in transit with TLS. At rest, data lives only in memory on encrypted nodes and is disposed with the session. In the self-hosted product, your library is sealed into an AES-256-GCM vault whose key is derived from your passphrase with argon2id.

Agents pay per request over x402. They call an endpoint, get an HTTP 402 with payment terms, pay in USDC on Base, and retry with the signed payload. No signup, no API key. Each paying wallet gets its own disposable sandbox.

Yes, through the KYC-verified persistent tier. Because durably storing user content carries legal responsibility, the persistent tier requires identity verification: you submit formation documents, tax ID, intended purpose, and data locale (for GDPR). That application is encrypted in your browser's request and stored only on an isolated review system — never on the cluster — and reviewed by a human within about three business days. Approved tenants get a private encrypted vault. Apply →

No. Instance sandboxes have all models baked in and no network egress; an outbound attempt is blocked and flagged for review. The gateway never logs request paths, queries, or bodies.

Yes. The same engine runs in Docker on a laptop or on your own Kubernetes cluster, and there is a fully offline Android app that makes no network calls at all. See the self-host guide.