hCaptcha Challenger

🚀 Gracefully face hCaptcha challenge with multimodal large language model.

Introduction

Does not rely on any Tampermonkey script.

Does not use any third-party anti-captcha services.

Just implement some interfaces to make AI vs AI possible.

Documentation: English | 简体中文 | Русский 🇷🇺 | Tiếng Việt 🙌

Modifications by Malvryx CodeLabs

This is a modified fork. The following changes were made by Malvryx CodeLabs on top of the upstream QIN2DIM/hcaptcha-challenger project.

• Added multiple alternative LLM providers alongside the default Gemini backend, selectable via a new LLM_PROVIDER setting on AgentConfig. The active provider only needs its own API key.
  • Groq (LLM_PROVIDER="groq") — Groq's vision models (Llama 4 Scout, default). New GroqProvider talking to Groq's OpenAI-compatible endpoint via httpx (no new required dependencies).
  • Gemini custom endpoint — GEMINI_BASE_URL routes the native Gemini provider through a proxy/gateway while keeping its schema-enforced output.
  • Generic OpenAI-compatible (LLM_PROVIDER="openai") — any /v1/chat/completions endpoint (e.g. Qwen-VL) via OPENAI_BASE_URL / OPENAI_API_KEY / OPENAI_MODEL.
  • Qwen via aikit.club (LLM_PROVIDER="aikit") — AikitProvider with automatic token refresh (/v1/refresh, proactive + on-401).
  • Robustness: auto-repair of malformed grid coordinates (e.g. weaker models returning ["00"] instead of [0,0]).
  • LLMProvider enum, provider model constants, and provider selection wired through models.py, the Reasoner base class, and RoboticArm.
  • Added example examples/demo_groq_agent.py, offline tests tests/test_provider_groq.py, and provider sections in the docs.

Gemini remains the default provider, so existing setups are unaffected.

What's features

Challenge Type	Pluggable Resource	Agent Capability
image_label_binary	ResNet ONNX classification #220401	✅
image_label_area_select: point	YOLOv8 ONNX detection #230826	✅
image_label_area_select: bounding box	YOLOv8 ONNX segmentation #230828	-
image_label_multiple_choice	ViT ONNX zero-shot motion #231109	-
image_drag_drop	Spatial Chain-of-Thought #250401	✅

Advanced Task	Pluggable Resource
Rank.Strategy	nested-model-zoo #231006
self-supervised challenge	CLIP-ViT #231022
Agentic Workflow	AIOps Multimodal Large language model #250331

Workflow

Tasks	Resource
ci: sentinel
ci: collector
datasets: VCS, annoate	#roboflow, #model-factory
model: ResNet - train / val
model: YOLOv8 - train / val
model: upload, upgrade	#objects, #modelhub
datasets: public, archive	#roboflow-universe, #captcha-datasets

Contributors

I would like to express my sincere gratitude to all the contributors.

What's next

• Dislock, the most advanced Discord Browser Generator. Powered by hCaptcha Solving AI.
• undetected-playwright, stash the fingerprint of playwright-based web agents.
• epic-awesome-gamer, gracefully claim weekly free games from Epic Store.

Reference

• Microsoft/playwright-python
• Anthropic/MCP
• Google/A2A
• Google/Gemini