How to Run Local LLMs with Claude Code

This step-by-step guide shows you how to connect open LLMs and APIs to Claude Code entirely locally, complete with screenshots. Run using any open model like Qwen3.5, DeepSeek and Gemma.

For this tutorial, we’ll use Qwen3.5 and GLM-4.7-Flash. Both are the strongest 35B MoE agentic & coding model as of Mar 2026 (which works great on a 24GB RAM/unified mem device) to autonomously fine-tune an LLM with Unsloth. You can swap in any other model, just update the model names in your scripts.

Qwen3.5 TutorialGLM-4.7-Flash TutorialClaude Code Tutorial

For model quants, we will utilize Unsloth Dynamic GGUFs to run any LLM quantized, while retaining as much accuracy as possible.

📖 LLM Setup Tutorials

Before we begin, we firstly need to complete setup for the specific model you're going to use. We use llama.cpp which is an open-source framework for running LLMs on your Mac, Linux, Windows etc. devices. Llama.cpp contains llama-server which allows you to serve and deploy LLMs efficiently. The model will be served on port 8001, with all agent tools routed through a single OpenAI-compatible endpoint.

Qwen3.5 Tutorial

We'll be using Qwen3.5-35B-A3B and specific settings for fast accurate coding tasks. If you don't have enough VRAM and want a smarter model, Qwen3.5-27B is a great choice, but it will be ~2x slower, or you can use other Qwen3.5 variants like 9B, 4B or 2B.

1

Install llama.cpp

We need to install llama.cpp to deploy/serve local LLMs to use in Claude Code etc. We follow the official build instructions for correct GPU bindings and maximum performance. Change -DGGML_CUDA=ON to -DGGML_CUDA=OFF if you don't have a GPU or just want CPU inference. For Apple Mac / Metal devices, set -DGGML_CUDA=OFF then continue as usual - Metal support is on by default.

Copy

apt-get update
apt-get install pciutils build-essential cmake curl libcurl4-openssl-dev git-all -y
git clone https://github.com/ggml-org/llama.cpp
cmake llama.cpp -B llama.cpp/build \
    -DBUILD_SHARED_LIBS=OFF -DGGML_CUDA=ON
cmake --build llama.cpp/build --config Release -j --clean-first --target llama-cli llama-mtmd-cli llama-server llama-gguf-split
cp llama.cpp/build/bin/llama-* llama.cpp

2

Download and use models locally

Download the model via huggingface_hub in Python (after installing via pip install huggingface_hub hf_transfer). We use the UD-Q4_K_XL quant for the best size/accuracy balance. You can find all Unsloth GGUF uploads in our Collection here. If downloads get stuck, see Hugging Face Hub, XET debugging

Copy

hf download unsloth/Qwen3.5-35B-A3B-GGUF \
    --local-dir unsloth/Qwen3.5-35B-A3B-GGUF \
    --include "*UD-Q4_K_XL*" # Use "*UD-Q2_K_XL*" for Dynamic 2bit

We used unsloth/Qwen3.5-35B-A3B-GGUF , but you can use another variant like 27B or any other model like unsloth/Qwen3-Coder-Next-GGUF.

3

Start the Llama-server

To deploy Qwen3.5 for agentic workloads, we use llama-server. We apply Qwen's recommended sampling parameters for thinking mode: temp 0.6, top_p 0.95 , top-k 20. Keep in mind these numbers change if you use non-thinking mode or other tasks.

Run this command in a new terminal (use tmux or open a new terminal). The below should fit perfectly in a 24GB GPU (RTX 4090) (uses 23GB) --fit on will also auto offload, but if you see bad performance, reduce --ctx-size .

We used --cache-type-k q8_0 --cache-type-v q8_0 for KV cache quantization for less VRAM usage. For full precision, use --cache-type-k bf16 --cache-type-v bf16 According to multiple reports, Qwen3.5 degrades accuracy with f16 KV cache, so do not use --cache-type-k f16 --cache-type-v f16 which is also on by default in llama.cpp. Note bf16 KV Cache might be slightly slower on some machines.

Copy

./llama.cpp/llama-server \
    --model unsloth/Qwen3.5-35B-A3B-GGUF/Qwen3.5-35B-A3B-UD-Q4_K_XL.gguf \
    --alias "unsloth/Qwen3.5-35B-A3B" \
    --temp 0.6 \
    --top-p 0.95 \
    --top-k 20 \
    --min-p 0.00 \
    --port 8001 \
    --kv-unified \
    --cache-type-k q8_0 --cache-type-v q8_0 \
    --flash-attn on --fit on \
    --ctx-size 131072 # change as required

You can also disable thinking for Qwen3.5 which can improve performance for agentic coding stuff. To disable thinking with llama.cpp add this to the llama-server command:

--chat-template-kwargs "{\"enable_thinking\": false}"

GLM-4.7-Flash Tutorial

1

Install llama.cpp

We need to install llama.cpp to deploy/serve local LLMs to use in Claude Code etc. We follow the official build instructions for correct GPU bindings and maximum performance. Change -DGGML_CUDA=ON to -DGGML_CUDA=OFF if you don't have a GPU or just want CPU inference. For Apple Mac / Metal devices, set -DGGML_CUDA=OFF then continue as usual - Metal support is on by default.

Copy

apt-get update
apt-get install pciutils build-essential cmake curl libcurl4-openssl-dev git-all -y
git clone https://github.com/ggml-org/llama.cpp
cmake llama.cpp -B llama.cpp/build \
    -DBUILD_SHARED_LIBS=OFF -DGGML_CUDA=ON
cmake --build llama.cpp/build --config Release -j --clean-first --target llama-cli llama-mtmd-cli llama-server llama-gguf-split
cp llama.cpp/build/bin/llama-* llama.cpp

2

Download and use models locally

Download the model via huggingface_hub in Python (after installing via pip install huggingface_hub hf_transfer). We use the UD-Q4_K_XL quant for the best size/accuracy balance. You can find all Unsloth GGUF uploads in our Collection here. If downloads get stuck, see Hugging Face Hub, XET debugging

We used unsloth/GLM-4.7-Flash-GGUF , but you can use anything like unsloth/Qwen3-Coder-Next-GGUF - see Qwen3-Coder-Next

Copy

import os
os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"
from huggingface_hub import snapshot_download

snapshot_download(
    repo_id = "unsloth/GLM-4.7-Flash-GGUF",
    local_dir = "unsloth/GLM-4.7-Flash-GGUF",
    allow_patterns = ["*UD-Q4_K_XL*"],
)

3

Start the Llama-server

To deploy GLM-4.7-Flash for agentic workloads, we use llama-server. We apply Z.ai's recommended sampling parameters (temp 1.0, top_p 0.95).

Run this command in a new terminal (use tmux or open a new terminal). The below should fit perfectly in a 24GB GPU (RTX 4090) (uses 23GB) --fit on will also auto offload, but if you see bad performance, reduce --ctx-size .

We used --cache-type-k q8_0 --cache-type-v q8_0 for KV cache quantization to reduce VRAM usage. If you see reduced quality, instead you can use bf16 but it will increase VRAM use by twice: --cache-type-k bf16 --cache-type-v bf16

Copy

./llama.cpp/llama-server \
    --model unsloth/GLM-4.7-Flash-GGUF/GLM-4.7-Flash-UD-Q4_K_XL.gguf \
    --alias "unsloth/GLM-4.7-Flash" \
    --temp 1.0 \
    --top-p 0.95 \
    --min-p 0.01 \
    --port 8001 \
    --kv-unified \
    --cache-type-k q8_0 --cache-type-v q8_0 \
    --flash-attn on --fit on \
    --batch-size 4096 --ubatch-size 1024 \
    --ctx-size 131072 #change as required

You can also disable thinking for GLM-4.7-Flash which can improve performance for agentic coding stuff. To disable thinking with llama.cpp add this to the llama-server command:

--chat-template-kwargs "{\"enable_thinking\": false}"

Claude Code Tutorial

Once you are done doing the first steps of setting up your local LLM, it's time to setup Claude Code. Claude Code is Anthropic's agentic coding tool that lives in your terminal, understands your codebase, and handles complex Git workflows via natural language.

Install Claude Code and run it locally

🕵️Fixing 90% slower inference in Claude Code

To solve this, edit ~/.claude/settings.json to include CLAUDE_CODE_ATTRIBUTION_HEADER and set it to 0 within "env"

For example do cat > ~/.claude/settings.json then add the below (when pasted, do ENTER then CTRL+D to save it). If you have a previous ~/.claude/settings.json file, just add "CLAUDE_CODE_ATTRIBUTION_HEADER" : "0" to the "env" section, and leave the rest of the settings file unchanged.

🌟Running Claude Code locally on Linux / Mac / Windows

Navigate to your project folder (mkdir project ; cd project) and run:

To use Qwen3.5-35B-A3B, simply change it to:

To set Claude Code to execute commands without any approvals do (BEWARE this will make Claude Code do and execute code however it likes without any approvals!)

Try this prompt to install and run a simple Unsloth finetune:

After waiting a bit, Unsloth will be installed in a venv via uv, and loaded up:

and finally you will see a successfully finetuned model with Unsloth!

IDE Extension (VS Code / Cursor)

You can also use Claude Code directly inside your editor via the official extension:

• Install for VS Code

• Install for Cursor

• Claude Code in VS Code docs

Alternatively, press Ctrl+Shift+X (Windows/Linux) or Cmd+Shift+X (Mac), search for Claude Code, and click Install.