# TRACE

> Privacy-first, stateless vector search and RAG. Reverse image search, cross-modal
> text-to-image, and document/video/audio search run in a stateless per-session
> sandbox that holds all data in memory and is destroyed the moment the session ends
> — no persistent index, nothing retained between sessions. People connect a wallet;
> AI agents pay per request over x402 (USDC on Base) with no account. TRACE is the
> first x402-native RAG and multimedia search engine, and a stateless / ephemeral /
> disposable alternative to hosted vector databases like Pinecone, Weaviate, and Qdrant.

TRACE is built on a sign-bit quantized vector engine: embeddings are compressed
to 1 bit per dimension, giving a ~32× smaller index than float vectors and ~36 ms
p50 search at 100k items, with search running as XOR + popcount (Hamming distance).
Every tenant session is an isolated gVisor sandbox with no network egress, no
disk writes, and no query logging; it is recycled on exit, idle, or TTL.

## Key facts
- Reverse image search and cross-modal text→image search (CLIP embeddings).
- Semantic search over documents, PDFs, video keyframes (seeks to the matching
  moment), and audio (transcript + sound embeddings).
- Two access rails: wallet sign-in for humans (ephemeral session tenant), x402
  pay-per-request for agents ($0.006/search, $0.03/ingest, no signup).
- Fully ephemeral by default: data is a duplicate held in RAM, deleted with the
  session. A KYC-verified persistent tier is available for durable storage.
- Self-hostable via Docker or Kubernetes; an offline Android app makes zero
  network calls.

## Docs
- [Quickstart](https://trace-cloud.aioniq.ai/docs/quickstart.html): sign in and search.
- [Product thesis (privacy & security)](https://trace-cloud.aioniq.ai/thesis.md): why TRACE is private by construction and disposable by default — threat model, use cases, tutorial.
- [agents.txt (machine-readable interaction + conduct)](https://trace-cloud.aioniq.ai/agents.txt): how to call TRACE and the agent code of conduct.
- [Agent guide (full)](https://trace-cloud.aioniq.ai/agents.md): why + how agents use TRACE for RAG + multimedia, with diagrams and use cases.
- [Agent API (x402)](https://trace-cloud.aioniq.ai/docs/agents.html): the 402→pay→retry
  cycle, Python and curl examples, endpoint prices, Bazaar discovery.
- [API reference](https://trace-cloud.aioniq.ai/docs/api.html): full endpoint shapes.
- [OpenAPI spec](https://trace-cloud.aioniq.ai/openapi.json): machine-readable API.
- [Self-hosting](https://trace-cloud.aioniq.ai/docs/self-host.html): Docker and k8s.
- [FAQ](https://trace-cloud.aioniq.ai/faq.html): disposal, encryption, persistence, agents.
- [Comparison](https://trace-cloud.aioniq.ai/vs.html): how TRACE differs from hosted
  vector databases and image-search APIs.

## For agents
The paid endpoints return HTTP 402 with Bazaar-compatible payment requirements in
`accepts[]`. Pay in USDC on Base (network eip155:8453) to the advertised wallet,
retry with the signed `X-PAYMENT` header, and receive results plus an
`X-PAYMENT-RESPONSE` settlement receipt. No API key or account is required.