Compress a query with one POST.

Endpoint

POST https://vrbose.ai/detokenate
Authorization: Bearer vb_live_...
Content-Type: application/json

{ "q": "the original query" }

Plain text body

With Content-Type: text/plain, send the raw query as the request body. A successful response is text/plain containing only the compressed query (no JSON, no usage block). Errors still return JSON.

POST https://vrbose.ai/detokenate
Authorization: Bearer vb_live_...
Content-Type: text/plain

the original query

Version-specific URL (same behavior today): POST https://vrbose.ai/detokenate/v1

Compression engines

Optional JSON field compression_engine or plain-text header X-Vrbose-Compression-Engine selects how Vrbose compresses your query. Values: auto (default), ironwings, spacy, or stilted.

• auto — tries Ironwings when configured, otherwise Stilted.
• ironwings — Rust sidecar (`IRONWINGS_URL`); falls back to Stilted on failure.
• spacy — grammar-aware sidecar (`SPACY_URL`); opt-in only; falls back to Stilted on failure.
• stilted — deterministic regex pass; always available offline.

JSON responses include compression_engine with the engine that produced q. Response headers X-Vrbose-Compression-Engine and, when applicable, X-Vrbose-Compression-Fallback-From mirror the same metadata.

Response

{
  "q": "compressed query",
  "compression_engine": "stilted",
  "usage": {
    "original_tokens_estimate": 1200,
    "compressed_tokens_estimate": 650,
    "estimated_savings_percent": 46
  }
}

Limits

Free accounts receive 20 successful API queries per day. Paid accounts are unlimited at $0.01 per successful billable query. Stripe invoices run when accumulated unpaid usage reaches $10; smaller balances roll forward. Requests are capped at 64 KB, and past-due paid accounts must update billing before additional API calls succeed.

Machine-readable contract

The OpenAPI spec is available at /openapi.json for client generation, audits, and agent integrations.