why run npm install before running docker image？

[BrownTan]

The command “npm install” exists in Dockerfile!
If I can't use npm and can only use it inside Docker due to no root, what should I do?

[mratsim]

The README is wrong

You should instead run docker build

[BrownTan]

The README is wrong

You should instead run docker build

Thank you, and I have solved this problem a few days ago.
I think the biggest pain point of the project is that some models do not support JSON Schema.

[mratsim]

Indeed, I tried with Gemma3, worked, but with QwQ, no.

Similar to what I posted here: jina-ai/node-DeepResearch#93 (comment)

Deepseek-R1 doesn't support tool calling at the moment in popular backends like vLLM

Hence libraries that required structured output like json cannot use it:

• [Feature]: support tool and reasoning together vllm-project/vllm#14429
• [BUG]why some models not call custom tools? crewAIInc/crewAI#2383
• https://huggingface.co/Qwen/QwQ-32B-Preview/discussions/52#67ca760d638303ce07dec500

[BrownTan]

Indeed, I tried with Gemma3, worked, but with QwQ, no.

Similar to what I posted here: jina-ai/node-DeepResearch#93 (comment)

Deepseek-R1 doesn't support tool calling at the moment in popular backends like vLLM
Hence libraries that required structured output like json cannot use it:

* https://github.com/vllm-project/vllm/issues/14429

* https://github.com/crewAIInc/crewAI/issues/2383

* https://huggingface.co/Qwen/QwQ-32B-Preview/discussions/52#67ca760d638303ce07dec500

Yes, I believe functionalities similar to DeepResearch are not yet ready for commercial use.