---
title: "Comment by Kartik N V J K on \"Local LLM Inference: What to Know Before You Start\" | Hashnode"
description: "The detail that matches my own experience is Claude Code hitting API-compatibility problems against a local server while plain chat worked. I ran into the same"
url: "https://hashnode.com/posts/local-llm-inference-what-to-know-before-you-start/6abb362687ee11aa2e1512ed/comment/6abc154cf115f8caf396b76a"
---

[Discussion](https://hashnode.com/posts/local-llm-inference-what-to-know-before-you-start/6abb362687ee11aa2e1512ed)

Comment [Local LLM Inference: What to Know Before You Start](https://hashnode.com/posts/local-llm-inference-what-to-know-before-you-start/6abb362687ee11aa2e1512ed)

[Kartik N V J K](https://hashnode.com/@kartik-nvj)

[Sep 29](https://hashnode.com/posts/local-llm-inference-what-to-know-before-you-start/6abb362687ee11aa2e1512ed/comment/6abc154cf115f8caf396b76a)

The detail that matches my own experience is Claude Code hitting API-compatibility problems against a local server while plain chat worked. I ran into the same wall: the multi-turn tool-call loop expects OpenAI-exact response shapes, and Lemonade's llama.cpp bridge drops fields the agent needs mid-task. Did switching to OpenCode fix the tool-call handshake for you, or did you still patch responses?

[Ikraj Singh](https://hashnode.com/@ikraj-singh)

[Sep 30](https://hashnode.com/posts/local-llm-inference-what-to-know-before-you-start/6abb362687ee11aa2e1512ed/comment/6abc154cf115f8caf396b76a#reply-6abc8d165b882b5f65c8872a)

Thanks for the read! I switched to OpenCode completely, it worked with all the models I tried. It is not extensive like other CLI tools I have used like claude code or codex, but its close.