DeepSeek

A research code repository for studying how fine-tuning can trigger verbatim recall of copyrighted books in large language models. It includes preprocessing, fine-tuning, generation, and memorization-evaluation scripts, with setup notes and example data.

A CLI proxy that lets Claude Code run against DeepSeek V4 Pro, OpenRouter, Fireworks AI, or Anthropic-compatible backends while keeping the same agent workflow.

An open-source project that deploys a persistent DeepSeek-based coding agent on Cloudflare Workers with one-command installation and durable workspace support.

DSCode is an open-source coding agent for working in real repositories from the terminal. It supports install via npm, connects to DeepSeek, and highlights local sessions, sandboxed commands, parallel agents, and cost tracking.

Windows desktop client for DeepSeek Harness that wraps the official web UI into a native app with installer, tray integration, built-in browser, and terminal support.

A PyTorch reference implementation of DeepSeek Sparse Attention from the LLMs-from-scratch project, including a lightweight indexer, top-K token selection, and a GPT-style model with KV-cache support.

A Python-based coordination protocol and daemon for running multiple LLM sessions in parallel with shared awareness, aimed at Claude Code and other Anthropic-compatible models.

An OpenAI-compatible API proxy for using DeepSeek models with standard OpenAI SDKs, billed pay-as-you-go and accessible without phone verification.

An MCP server that converts images into ASCII art so text-only LLMs can attempt image interpretation. The README includes a short explanation, example setup, and a clear description of its limitations.

Retriever AI’s blog post explains a browser-agent architecture built around DeepSeek Flash, text-only DOM handling, and executable code as the planning layer. It includes concrete examples of the harness, the model’s role, and the cost argument behind the approach.

A DeepSeek paper about DSpark, a full-stack codebase for training and evaluating speculative decoding algorithms to speed up LLM inference.