NeuralDeep — an AI hub for developers

What it is: a unified OpenAI-compatible API to large language models (LLMs) running on our own GPUs. It's an LLM proxy/gateway and platform — NOT an image generator and NOT a 3D service. One sk- key, base URL https://api.neuraldeep.ru/v1, compatible with the OpenAI SDK, opencode, Cursor, Cline, Continue and any OpenAI client.

How it differs from token resellers: the models run on our own hardware (not a resold third-party API), and you pay with a monthly subscription instead of a metered per-token bill.

Models

• DaisyGPT (based on Qwen 3.6, 35B-A3B) — chat, tools, reasoning, vision, 256k context
• GPT-OSS 120B — chat, tools, reasoning, 131k context
• Gemma 4 31B — chat, multimodal, 262k context
• Kimi K2.6 — external frontier model (paid tiers)
• Embeddings: e5-large, bge-m3, jina-v4, GigaEmbeddings · Rerank: bge-reranker · Whisper: audio transcription

On the free tier only open models are available — DaisyGPT (Qwen 3.6) and GPT-OSS 120B. Frontier models (GPT-5.2, Claude, Gemini, DeepSeek, Kimi) are on paid tiers.

Platform services

• LLM API — chat, embeddings, rerank, transcription, vision
• Drift — a personal AI agent with memory, a code sandbox and Telegram
• Search API — web and Telegram search, crawling
• OCR API — PDF/scan → markdown
• Image API — generation/upscale/background removal

Links

• Documentation: hub.neuraldeep.tech/docs
• Reference for AI agents (markdown): neuraldeep.tech/llms-full.txt
• API: api.neuraldeep.ru/v1 · Dashboard: hub.neuraldeep.tech/app