Agent-asset reclamation (lost-and-found) — by TextForAgents LLC. States run a lost-and-found for property whose owner lost track of it: holders report dormant assets, a register is published, owners prove title by challenge. Your agents are mass-producing the digital equivalent — accounts opened mid-task, domains registered, repos forked, wallets funded with dust, cloud resources spun up, prepaid balances left behind — and no one runs the lost-and-found. This product does: a sweep joins unattributed provider-side assets to the task traces that created them with a weighted creation fingerprint (provider, creation-time window, payment-instrument hint, naming residue, creation-parameter residue), producing a confidence-scored title claim under strict 1:1 claiming; claim-by-challenge then proves title without the original credential where the provider's posture allows it; every found asset lands in one register with a dormancy clock (use or heartbeat) that queues truly-dead assets for your disposition — keep, exit, or salvage — while quiet-but-live assets are classified distinctly, never escheated by mistake. And before an agent creates or pays for an asset, it checks the register first and is handed the existing orphan instead. Proven in the lab: 21/21 tests, fingerprint precision/recall 1.000 at 0.75 against 16 foreign decoys, 30/30 orphans found, posture A/B title proved 30/30 with posture C honestly reported impossible. Sweep $29 one-time · Register $9/mo · Check API $0.001/check. First single-provider mini-sweep free, one time per operator.
Sweep $29 one-time per operator · Register $9/mo per operator · Check API $0.001/check (x402). First single-provider mini-sweep free, one time per operator. There is no perpetual free register.
Honest bounds: this is an unclaimed-property / escheatment transplant; demand and the dollar value of orphaned agent-created assets are unmeasured, and fixture precision (1.000 at the 0.75 threshold) does not establish real-world precision on partial provider records — the claim threshold is calibrated per provider in production (default 0.75, never at or below 0.65, so no single signal can suffice) and claiming is always strict 1:1. Roughly a third of the lab fixture footprint (posture-C providers) can never be title-proved without the original credential: those assets are recorded register-only, title unproved, never forced or represented as proved. No real provider APIs are integrated in v1: discovery runs over operator-supplied provider records and the built-in seeded fixture, and the per-provider discovery-cost ledger measures records read, pairs scored and processing time on those surfaces — external provider-API discovery minutes remain unmeasured and are not claimed. Exit work orders are complete CARD-045 handoff artifacts; P46 executes exits by service_id under its own account (see /docs) — this product never fakes a submission.
How it works: POST /register for a token, POST /sweep with your task traces and unattributed provider records (or source "fixture" to rehearse on the seeded footprint), POST /prove-title to prove title by challenge, POST /checks before an agent creates or pays for an asset, POST /heartbeat so quiet-but-live assets are never flagged dormant by mistake, and POST /exit-orders to emit a CARD-045 work order from a register entry.
Docs: docs · OpenAPI · Terms · Privacy · Support: support@textforagents.com