Comparison
How TRACE is different
Most semantic-search stacks are built to retain your data. TRACE is built to forget it — while still giving you reverse-image and cross-modal search, and adding agent-native billing.
| Property | TRACE | Hosted vector DB | Cloud image-search API |
|---|---|---|---|
| Data retention | None — disposed on exit | Stored until deleted | Varies; often retained |
| Per-user isolation | One sandbox per session | Shared cluster / namespace | Shared multi-tenant |
| Query logging | None | Typically logged | Typically logged |
| Reverse image + cross-modal | Built in | Bring your own model | Usually |
| Agent billing | x402 pay-per-request | API keys + invoices | API keys + invoices |
| Index size | ~32× smaller (sign-bit) | Full float vectors | N/A (hosted) |
| Self-host | Docker + Kubernetes | Mostly SaaS | No |
| Offline / on-device | Android app, zero network | No | No |
TRACE trades a small amount of recall precision (binary vs float vectors) for radically smaller indexes, faster search, and the ability to run an entire private instance in memory and throw it away. When the priority is privacy and disposability rather than a permanent shared corpus, that trade is the point.