A self-hosted captcha solver that clones the YesCaptcha API

What it does OhMyCaptcha is a FastAPI service that solves captchas by driving browsers with Playwright and querying vision models. It mimics the YesCaptcha API—createTask, getTaskResult, getBalance—so existing clients can swap it in as a self-hosted replacement for commercial solvers. It covers 19 task types, from reCAPTCHA v2/v3 and hCaptcha to image-to-text and grid-classification challenges.

The interesting bit The project splits cognitive load across two models: a lightweight local multimodal model (Qwen3.5-2B via SGLang) handles image analysis, while a separate cloud model tackles trickier reasoning. Browser automation and vision inference are wrapped in a single async queue that lives entirely in memory, with tasks evaporating after ten minutes.

Key highlights

• Drop-in YesCaptcha compatibility: same createTask/getTaskResult/getBalance protocol
• 19 task types: browser automation for reCAPTCHA, hCaptcha, Cloudflare Turnstile, plus image classification and OCR-style recognition
• Dual-model backend: local SGLang or vLLM for images, plus any OpenAI-compatible cloud endpoint for harder tasks
• Deploys to Render, Hugging Face Spaces, or local Docker with headless Chromium
• In-memory async task queue with automatic retries

Caveats

• Tasks are stored in memory with a 10-minute TTL; there is no persistence layer
• Browser-based solving depends heavily on environment, IP reputation, and target-site behavior
• minScore is accepted for compatibility but not actually enforced

Verdict A practical DIY alternative if you want a self-hosted YesCaptcha-compatible service and are comfortable managing Playwright and local LLM inference. Look elsewhere if you need the reliability guarantees and full feature parity of a commercial solver.