dummy-llm fakes OpenAI, Anthropic, Gemini and Ollama: a fake LLM server

⚡ The gist in 5 seconds - dummy-llm has been published on GitHub — a fake LLM server that echoes back the text you send it or returns a response based on your own template. - It supports streaming in the real OpenAI, Anthropic, Gemini and Ollama protocols. - Behavior is deterministic: the same input always produces the same output. ### 🔍 What was found The lco117/dummy-llm project has been posted on GitHub as an open-source fake LLM server. According to the repository description, it can do two things: echo responses — returning exactly what you send it — or respond using your own custom template. The key feature is streaming compatibility with four real protocols at once: OpenAI, Anthropic, Gemini and Ollama. In other words, a client written for any of these APIs sees a familiar response format. The mechanics are simple and predictable: the server doesn't run any model, it merely imitates its interface. The output is deterministic — with the same request, you're guaranteed to get the same response, which is what sets dummy-llm apart from real LLMs with their temperature settings and random token sampling. ### 💡 Why it matters For developers building wrappers around LLM providers, one thing always gets in the way: real APIs require keys, cost money per token, and produce unstable responses that make tests flaky. dummy-llm fills this niche — it provides a stable response in a familiar protocol without touching a real model or the network. It's handy for unit tests, local streaming debugging, and verifying response handling in pipelines, especially when you need to reproduce the same scenario in CI.