CheapAIS DeepSeek H

DeepSeek H is an agent harness built around open-source DeepSeek models, DeepSeek-R1 for reasoning and DeepSeek-V3 for general generation, so you get a capable, self-hosted AI agent without paying per-token API fees. Ollama serves the models locally, Open WebUI gives you a private interface to talk to them, and the harness connects DeepSeek to workflow automation so you can run reasoning and agentic tasks entirely on your own server. 

 Because DeepSeek H runs on CPU-only hardware, it serves requests in batch or async mode rather than streaming back in real time. That is the honest trade: it excels at offline inference, RAG, document processing, and queues of jobs you are not actively watching, but it is not for latency-sensitive response-time chat. If you want a model that reasons carefully over a long task, this is a strong fit.

 DeepSeek-R1 / DeepSeek-V3 — open reasoning and generation models, served locally via Ollama

 Open WebUI — private, ChatGPT-like interface to your model and agents

 n8n — workflow automation to wire DeepSeek into your pipelines

 Supabase — database, vector store, and auth layer for agent memory and RAG

 Ollama — cross-platform platform to pull and run local DeepSeek models

 Caddy — automated HTTPS/TLS for custom domains

 The stack ships pre-installed on a CheapAIS server with full root access and SSH, so you own the hardware, the models, and the data.