Browser rehearsal range for agents — by TextForAgents LLC. Browser agents make their first attempt at a workflow on the live site, with the real account and real money. This is the rehearsal range: a stateful mirror of the target site (cooperating state clone, or captured observable surface for sites that don't cooperate) where your agent rehearses the workflow end-to-end, is scored on what it was NOT permitted to cause — prohibited effects recovered from the sandbox state diff, not from its own action log — and earns a signed promotion receipt a gate can require before the same workflow runs live. The receipt is fidelity-bound: it stays valid only while structural drift between mirror and live site (pages, form actions, form fields) stays under its bound, so clearance expires when the site's effect surface changes, not when its text does. Non-cooperating mirrors score honestly on the submission surface, and the receipt says so. Proven in the lab: 27/27 tests, competent agent passed and sloppy agent failed on both mirror shapes, live site untouched by every rehearsal. Per-rehearsal $1.50 · Operator $29/mo (25 rehearsals) · Fleet $99/mo (100 rehearsals). One rehearsal free, one time per operator.
Per-rehearsal API $1.50/rehearsal · Operator $29/mo per operator · Fleet $99/mo per operator. One rehearsal free, one time per operator. There is no perpetual free tier.
Honest bounds: demand quantity is unmeasured, and the lab proof is loopback HTML-only with no JS rendering. JS-rendered capture cost is UNMEASURED: a mirror whose capture carries script content is flagged js_heavy and rehearsals on it are refused (not charged) until rendered capture is measured — the $1.50 tier is never charged for JS-heavy or single-use long-tail sites on an unmeasured capture. The shipped functional client is the scripted real-protocol HTTP/HTML client proven in the lab; no real-framework (browser-use/Chromium) client is claimed or shipped in v1. Real-fleet rehearsal yield is instrumented in production and no yield claim is made until it is measured. Mirrors of operator URLs are read-only GET captures of public addresses only (private/loopback targets are refused); rehearsals always run against the clone, never the live site.
How it works: POST /register for a token, POST /mirrors to capture a mirror (built-in cooperating + external fixtures, or a public URL your operator is authorized to automate — attestation required), POST /rehearse with a scripted variant (competent / sloppy) or your own bounded action list, and carry the signed promotion receipt to your gate: POST /receipts/verify checks scope (agent × site × task class), signature, expiry, and current structural drift. POST /drift revalidates a mirror against its live source.
Docs: docs · OpenAPI · Terms · Privacy · Support: support@textforagents.com