I made a free DeepSeek coding agent using Gemini CLI called DeepGem (github.com)

DeepGem is an open-source terminal agent (deepgem) that routes prompts between DeepSeek (via its OpenAI‑compatible API) and Google’s Gemini CLI, letting you mix fast chat, deliberative reasoning, and agentic tool use from a single CLI. It provides simple commands (deepgem, dg, deepgem chat, deepgem gem) and a one‑click quickinstall.bat that auto-installs the package, adds commands to PATH, handles Windows UTF‑8 and npm quirks, and can install the Gemini CLI. Authentication supports DeepSeek API keys and either Gemini OAuth or API keys. DeepSeek exposes two model IDs (deepseek-chat and deepseek-reasoner) with streaming enabled by default; the built‑in router sends code/tooling prompts to Gemini (shell/file ops, web fetch/search, MCP), long “think step‑by‑step” prompts to deepseek-reasoner, and everything else to deepseek-chat. For developers and the AI/ML community this is notable because it bundles model-driven reasoning with tool-enabled agentic actions in a lightweight, reproducible CLI workflow—use cases include code analysis, automated refactors, test generation, and PR drafting directly from a repo. Technical implications include easy local integration (Python 3.10+, Node/npm), pip/pipx installation, Docker support, env vars for configuration (DEEPSEEK_API_KEY, GEMINI_API_KEY, DEEPGEM_DEFAULT_GEMINI_MODEL), and a smart router that reduces manual model/tool selection. The project is MIT‑licensed and designed to simplify building practical, agentic developer tooling using both DeepSeek’s reasoning and Gemini’s tool access.