Hi Simon 👋
I've been using llm for a while — the plugin architecture is genuinely well-designed.
The ecosystem has individual plugins for DeepSeek, Qwen, Kimi, etc., but each needs its own API key and registration. I'd like to contribute an llm-anyllm plugin — a single API key that gives access to 24+ Chinese and international models (no Chinese phone number or Alipay needed):
1.llm keys set anyllm
2.llm -m anyllm/deepseek-v4-flash "Summarize this article"
3.llm -m anyllm/gemini-3.5-flash "Describe this image" -a image.jpg
4.llm -m anyllm/gpt-5.4-nano "Quick question"
This complements the existing single-model plugins — users who only need DeepSeek can still use llm-deepseek. AnyLLM is for those who want many models without registering on each platform.
Before opening a PR:
Would this be welcome in the plugin directory?
Any preference on model ID namespacing? I'm planning anyllm/deepseek-v4-flash etc. to avoid collisions.
For multimodal models — advanced plugin pattern with attachment support, or text-only first?
I'll follow the plugin tutorial docs, submit with tests (pytest, ruff, mypy, black), publish to PyPI, and maintain it long-term.
Thanks for building llm — it's my go-to CLI for LLMs.
— Leo Bennett [leo.indiedev@gmail.com])
Hi Simon 👋
I've been using llm for a while — the plugin architecture is genuinely well-designed.
The ecosystem has individual plugins for DeepSeek, Qwen, Kimi, etc., but each needs its own API key and registration. I'd like to contribute an llm-anyllm plugin — a single API key that gives access to 24+ Chinese and international models (no Chinese phone number or Alipay needed):
1.llm keys set anyllm
2.llm -m anyllm/deepseek-v4-flash "Summarize this article"
3.llm -m anyllm/gemini-3.5-flash "Describe this image" -a image.jpg
4.llm -m anyllm/gpt-5.4-nano "Quick question"
This complements the existing single-model plugins — users who only need DeepSeek can still use llm-deepseek. AnyLLM is for those who want many models without registering on each platform.
Before opening a PR:
Would this be welcome in the plugin directory?
Any preference on model ID namespacing? I'm planning anyllm/deepseek-v4-flash etc. to avoid collisions.
For multimodal models — advanced plugin pattern with attachment support, or text-only first?
I'll follow the plugin tutorial docs, submit with tests (pytest, ruff, mypy, black), publish to PyPI, and maintain it long-term.
Thanks for building llm — it's my go-to CLI for LLMs.
— Leo Bennett [leo.indiedev@gmail.com])