# Unslop API

> English writing checks and small, constrained edits. Most checks use no AI. Agents can register over HTTP; a human funds prepaid credits through Stripe when checkout is active.

Operator: 100Apps.Studio
Base URL: https://unslopapi.vercel.app
Read /pricing.json before onboarding: billing and checkout availability are separate flags. Never promise a working checkout if checkout_available is false.

## Start here
- [Agent quickstart](https://unslopapi.vercel.app/agents.html): registration, credentials, budgets, retries and human checkout.
- [OpenAPI 3.1](https://unslopapi.vercel.app/openapi.json): current request schemas and operations.
- [Pricing JSON](https://unslopapi.vercel.app/pricing.json): live mode, rates and credit packs.
- [Pricing and data handling](https://unslopapi.vercel.app/pricing.html): fees, failure charges and response caching.
- [API guide](https://unslopapi.vercel.app/api-guide.html): examples and response fields.
- [How checks work](https://unslopapi.vercel.app/how-it-works.html): deterministic diagnostics and limitations.
- [Research](https://unslopapi.vercel.app/references.html): studies and references.

## HTTP workflow
1. With your operator's authorization, POST /agent/register with username, password and agree:true. Save the returned API key and recovery code privately. No cookies, browser, email or CAPTCHA required.
2. Authenticate /v1 calls with Authorization: Bearer <key>. GET /v1/billing/balance and /v1/usage show funds and quotas.
3. POST /v1/billing/checkout with amount_usd:10, 25 or 50 and an Idempotency-Key. Give checkout_url to the human payer. Do not enter payment details or approve checkout for them. Poll balance at sensible intervals after they finish.
4. POST /v1/billing/quote for a conservative rewrite upper bound. This is not a guaranteed exact charge or a reservation.
5. Paid POST/PUT/DELETE /v1 requests need a unique Idempotency-Key and X-Max-Cost-USD decimal, 0.005–10. Reserve no more than the human-approved budget. Keep the exact key, body, path, profile revision and budget for retries.
6. POST /v1/analyze combines deterministic checks; POST /v1/rewrite scores first and edits small spans only when its configurable style gate passes. Confirm accepted and inspect validation before replacing a file.

## Constraints
- Style scores are heuristics, not proof of AI authorship. No claim of removing watermarks or guaranteeing undetectability.
- English only. Source text max 20,000 UTF-16 units; request body max 256,000 bytes. Paid serialized model inputs including guidance and schema max 128,000 UTF-8 bytes.
- Default rewrite: mixed OpenAI and Anthropic, 4 calls, 8 words / 60 UTF-16 units per patch, 15% change budget, gate score 20 and at least 50 words. Limits and models are in /v1/capabilities.
- Unknown model prices, Jev and external detectors fail closed for prepaid users. Owner keys are self-funded and retain existing access.
- 402: stop spending, check balance/budget, request human funding if needed. 409 request_in_progress: wait and retry the SAME request/key; do not create a second job. 410: cached result expired, no rerun. 429: respect the minute/day reset. 502/503: retry identical request/key to retrieve the stored outcome, not to start another paid attempt.
- Idempotency lasts 24 hours. Up to 5 recent response bodies remain available for up to 1 hour; expiry cleanup happens on the next paid request. A stalled request keeps its hold until owner reconciliation. Responses may contain your text; billing history has no input drafts.
- POST /agent/token with username/password rotates the API key AND recovery code, useful after a lost signup response. It is not a read-only login. No secrets in URLs, logs or public files.