Evaggelos Balaskas - System Engineer

Run OpenCode, an AI coding agent on your own machine — no cloud, no API, no data ever leaving your computer privacy first and no costs!

Introduction

If you’ve been curious about running AI coding agents entirely on your own machine then this blog post is for you. We will walk through setting up OpenCode, a terminal-based AI coding agent, and connecting it to LM Studio so it uses our local language models (LLMs) that you control.

What You’ll Need

Before we begin, make sure you have the following:

• A reasonably modern computer (macbook M series Pro with Apple Silicon work great, for this blog post I am using Macbook M4 Pro)
• LM Studio installed — download it from lmstudio.ai
• Additional you can install/use Visual Studio Code!

What is a AI coding Agent ?

so OpenCode is an open source AI coding agent that

• Turn Ideas into Real Tools
• Automate Boring Repetitive Tasks
• Fix Broken Things
• Connect Different Apps Together
• Explain Technical Jargon

eg.

I need a simple website for my dog-walking business where people can book a time and see my prices.

and opencode starts working on that

and the result is something like that, without writing a single line of code !

and yes, this example was made entirely on my macbook with opencode and lmstudio.

Install opencode

Open your terminal and run the official install script:

curl -fsSL https://opencode.ai/install | bash

or via brew (my preferable way)

brew install anomalyco/tap/opencode

This downloads and installs the opencode CLI tool. Once it’s done, close and reopen your terminal (or run source ~/.bashrc / source ~/.zshrc) so the command is available.

Verify it worked:

opencode --version

eg.

❯ opencode --version
1.2.27

Download a Model in LM Studio

Open LM Studio and use the built-in model browser to download a model. For this guide, we’ll use two good options that run well on consumer hardware:

• Ministral 3B — fast and lightweight, great for quick tasks
• Qwen 3.5 9B — more capable, needs more RAM/VRAM

Search for either model in LM Studio’s Discover tab and download it. Once downloaded, you’ll see it listed in your local models.

you can also use the CLI to get the models

eg. lms get mistralai/ministral-3-3b

❯ lms get mistralai/ministral-3-3b
   ✓ Satisfied mistralai/ministral-3-3b
   └─ ✓ Satisfied Ministral 3 3B Instruct 2512 Q4_K_M [GGUF]

⠋ Resolving download plan...

and list them lms ls

You have 3 models, taking up 9.62 GB of disk space.

LLM                                     PARAMS    ARCH        SIZE       DEVICE
mistralai/ministral-3-3b (1 variant)    3B        mistral3    2.99 GB    Local
qwen/qwen3.5-9b (1 variant)             9B        qwen35      6.55 GB    Local     

EMBEDDING                               PARAMS    ARCH          SIZE        DEVICE
text-embedding-nomic-embed-text-v1.5              Nomic BERT    84.11 MB    Local     

I am not going to analyse the models but in short, Qwen3.5-9B is best for a local, open, multimodal assistant that can handle:

• coding
• tool calling / agents
• long documents
• multilingual tasks
• document and image understanding

and fits in a a MacBook M4 Pro with 48GB RAM.

Important: Context Length

In simple words, context length is the AI’s short-term memory limit. Depending on the model and use, you need to adjust it on LM Studio. It is measured by tokens. Tokens are a chunk of a words. When using cloud AI models via API, the cost is measured on how many tokens you are using in a specific amount of time.

• Use Small Context Lenght (4096 - 8192) when you have a quick question, review/reply to a short email or debug a small snippet of code. It will produce a quick reply.

• Use Medium Context Length (32k) when you want to analyze a report, write a short story or working with a few coding files. It may take a couple minutes.

• Use Large Context Length (128+) when you want to upload a big document, or you want to analyze a project at once. It will be slow, slower on local machines.

See below details about LM Studio and LLM.

Start the LM Studio Local Server

LM Studio includes a built-in local API server that speaks the OpenAI API format — which means tools like opencode can talk to it directly.

In LM Studio, go to the Local Server tab (the <-> icon on the left sidebar) and click Start Server. By default it runs at http://localhost:1234.

Tweak Settings

to get the best from LM Studio

You can leave the server running in the background while you use opencode.

or you can use CLI to start LM Studio server:

❯ lms server start -p 1234 --bind 127.0.0.1
Waking up LM Studio service...
Success! Server is now running on port 1234

verify which models are available

by running in CLI a simple curl command curl -s http://localhost:1234/v1/models | jq .

{
  "data": [
    {
      "id": "qwen/qwen3.5-9b",
      "object": "model",
      "owned_by": "organization_owner"
    },
    {
      "id": "mistralai/ministral-3-3b",
      "object": "model",
      "owned_by": "organization_owner"
    },
    {
      "id": "text-embedding-nomic-embed-text-v1.5",
      "object": "model",
      "owned_by": "organization_owner"
    }
  ],
  "object": "list"
}

Configure opencode

opencode uses a config file called opencode.json stored in ~/.config/opencode/. You’ll need to create or edit this file to tell opencode about your LM Studio models.

Create the directory if it doesn’t exist:

mkdir -p ~/.config/opencode

Then create (or edit) the config file:

vim ~/.config/opencode/opencode.json

Paste in the following configuration:

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "lmstudio": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "lmstudio",
      "options": {
        "baseURL": "http://127.0.0.1:1234/v1",
        "apiKey": "lmstudio"
      },
      "models": {
        "qwen/qwen3.5-9b": {
          "name": "qwen3.5"
        },
        "mistralai/ministral-3-3b": {
          "name": "ministral3"
        }
      }
    }
  }
}

A few things to note:

• The baseURL points to LM Studio’s local server — keep this as-is unless you’ve changed LM Studio’s port.
• The apiKey value "lmstudio" is a placeholder — LM Studio doesn’t actually require a real API key, but the field needs to be present.
• The model IDs (e.g. mistralai/ministral-3-3b) must match exactly what LM Studio uses. You can check the model identifier in LM Studio’s model list.

Save and close the file.

Load a Model via the CLI (Optional but Useful)

LM Studio comes with a CLI tool called lms that lets you load and unload models from the terminal without opening the GUI. This is handy for scripting or keeping things lightweight.

First, unload any currently loaded model (to free memory):

lms unload "mistralai/ministral-3-3b"

Then load it fresh with a specific context window size:

lms load "mistralai/ministral-3-3b" --context-length 16384

The --context-length flag controls how much text the model can hold in memory at once. 16384 (16K tokens) is a good balance of capability and memory use. If you have more RAM to spare, try 32768.

full example with ministral

❯ lms unload "mistralai/ministral-3-3b"
Model "mistralai/ministral-3-3b" unloaded.

~
❯ lms load "mistralai/ministral-3-3b" --context-length 16384

Model loaded successfully in 2.67s.
(2.78 GiB)
To use the model in the API/SDK, use the identifier "mistralai/ministral-3-3b".

Test opencode with Your Local Model

opencode run --model lmstudio/mistralai/ministral-3-3b "capital of greece?"

The --model flag follows the format lmstudio/<model-id>, where the model ID matches what you put in the config file.

You should see the model respond directly in your terminal. If everything is connected correctly, the response comes from your local machine — no internet required.

output:

> build · mistralai/ministral-3-3b

Athens.

Run opencode with Your Local Model

Now you’re ready to use opencode on your project.

Change to the code directory cd project

and to start an interactive session in your current project directory, just run:

opencode

opencode will open its TUI (terminal user interface) where you can have a longer back-and-forth conversation, ask it to read files, write code, and more.

Verify opencode is using the correct model and type

/init

To initial your project. It will create an AGENTS.md file for your project.

or you can use VS code with the opencode extension and use it from there !

That’s it!

Happy coding my friends.

I can use apprise in Home Assistant and in my scripts and got signal, slack, email and many more working perfectly!

How I Built a Unified Notification Hub Using Apprise and Signal API

The Problem: Notification Chaos

If you’re like me, you probably have alerts coming from everywhere:

• Docker containers need to notify you when they crash
• Home Assistant wants to tell you when the door opens
• Monitoring scripts need to report when disk space is low
• Your Jellyfin server should alert you when transcoding fails

The problem? Each service wants to send notifications differently. Some support email, others want webhooks, a few can do Slack, and almost none support Signal natively.

Enter Apprise — the notification abstraction layer that changed how I handle alerts forever.

What is Apprise?

Apprise is a Python library (and API) that supports dozens of notification services through a single, unified interface. Think of it as a universal translator for notifications.

As you can see in the diagram above, Apprise acts as a notification router sitting between your services and your notification targets:

Input Sources:

• Docker Apps (Jellyfin, Nextcloud, etc.)
• Home Assistant
• Custom Scripts & Monitoring Tools
• Automation Platforms (n8n, Node-RED)

Output Targets:

• 📱 ntfy (Mobile Alerts)
• 💬 Slack
• 📧 Email
• 🔗 Webhooks
• 📞 Signal (via signal-cli)
• 📱 WhatsApp (Business API)
• 🔐 Threema & Viber

And that’s just a subset — Apprise supports 120+ notification services!

The Setup: Apprise API + Signal

In this guide, I’ll show you how to set up:

1. Apprise API — A REST API server for managing and sending notifications
2. Signal CLI REST API — A bridge to send messages via Signal
3. Integration — Connecting them so you can send Signal messages through Apprise

Why Signal?

Signal offers end-to-end encryption, privacy-focused design, and most importantly — it’s free for personal use. Perfect for receiving important alerts without relying on corporate platforms.

Step 1: Deploy Apprise API

Create a docker-compose.yml file for Apprise:

I’ve selected TCP Port 8800 as I am already using 8000 on my homelab.

services:
  apprise-api:
    image: caronc/apprise:latest
    container_name: apprise-api
    restart: unless-stopped
    ports:
      - "8800:8000"
    environment:
      - APPRISE_STATEFUL_MODE=simple
      - APPRISE_WORKER_COUNT=1
      - APPRISE_WORKER_OVERFLOW=10
      - LOG_LEVEL=info
    volumes:
      - ./apprise/config:/config
      - ./apprise/plugins:/plugin

Key Configuration Explained:

Start the service:

docker-compose up -d

Apprise API will now be available at http://localhost:8800

Step 2: Deploy Signal CLI REST API

Signal doesn’t have a native API, but the community has created bridges. We’ll use signal-cli-rest-api:

again, I am using TCP Port 9922 as I am using 8080 on my homelab already.

services:
  signal-api:
    image: bbernhard/signal-cli-rest-api
    container_name: signal-api
    environment:
      - MODE=native
    volumes:
      - ./signalcli-data:/home/.local/share/signal-cli
    ports:
      - "9922:8080"
    restart: unless-stopped

Important: The MODE=native setting uses the native Signal protocol library (libsignal) for better compatibility.

Start the service:

docker-compose up -d

PS. You can merge the two docker compose services into one if you prefer.

Step 3: Link Your Signal Account

Now for the magic — linking your Signal account to the API.

Open your browser and navigate to:

http://localhost:9922/v1/qrcodelink?device_name=signal-api

This will produce a QR image, that you need to scan with your mobile Signal app.

You’ll see a QR code (similar to the one above, but not blurred).

To link:

1. Open Signal on your phone (Android or iOS)
2. Go to Settings → Linked Devices
3. Tap the + button to add a new device
4. Scan the QR code

Once linked, your Signal account is now accessible via REST API!

Step 4: Register a Phone Number

Before sending messages, you need to register your phone number with Signal CLI. This is typically done automatically when you link the device, but if needed:

# Check if your number is registered
curl http://localhost:9922/v1/about

Step 5: Add Signal to Apprise

Now we connect Signal to Apprise. Apprise uses URL schemes to represent notification targets.

The Signal URL format is:

signal://<signal-api-host>:<port>/<recipient-phone-number>

Add your Signal configuration to Apprise:

curl -s -X POST http://127.0.0.1:8800/add/signal
  -d "urls=signal://localhost:9922/+306970000xyz"

Replace +306970000xyz with your phone number!

Expected response:

Successfully saved configuration

Note: Replace +306970000xyz with your actual phone number (in international format with + prefix).

Step 6: Send Your First Notification

Option A: Using the Apprise CLI

apprise -vv -t "Test Message Title" -b "Test Message Body"
   "signal://localhost:9922/+306970000xyz"

Option B: Using the Apprise API

curl -X POST http://localhost:8800/notify
  -H "Content-Type: application/json"
  -d '{
    "title": "Test Message Title",
    "body": "Test Message Body",
    "tag": "signal"
  }'

Option C: Send to Multiple Services at Once

Here’s where Apprise shines — send the same message to Signal, Slack, and email with one command:

apprise -t "Server Alert" -b "Disk usage at 90%"
  "signal://localhost:9922/+306970000xyz"
  "slack://token-a/token-b/token-c"
  "mailto://user:pass@gmail.com"

The Result

As you can see, the test message arrived successfully in Signal with both the title and body intact. This message was sent programmatically through the Apprise → Signal pipeline!

Real-World Use Cases

Now that you have the infrastructure set up, here are some practical applications:

Home Assistant Notifications

Add the below notify setup to your configuration.yaml:

# ebal, Sun, 15 Mar 2026 21:06:52 +0200
notify:
  - name: signal_notify
    resource: http://localhost:8800/notify
    url: "signal://localhost:9922/+306970000xyz"
    platform: apprise

and create a new Automation

1. Go to Settings → Automations & scenes
2. Tap the + button to create a new automation
3. Copy yaml code and replace your Device and Entity ID.

alias: Fridge Door Open
description: Send a message through Signal when Fridge Door is opened for more than 5sec
triggers:
  - type: opened
    device_id: <device_id>
    entity_id: <entity_id>
    domain: binary_sensor
    trigger: device
    for:
      hours: 0
      minutes: 0
      seconds: 5
conditions: []
actions:
  - action: notify.signal_notify
    metadata: {}
    data:
      message: Fridge Door Open for more than 5sec
      title: HomeAssistant Alert
mode: single

Docker Container Monitoring

# In your monitoring script
docker ps --format "{{.Names}}" | while read container; do
  if [ "$(docker inspect -f '{{.State.Running}}' $container)" != "true" ]; then
    apprise -t "Container Down" -b "$container has stopped"
      "signal://localhost:9922/+306970000xyz"
  fi
done

Automated Backup Alerts

#!/bin/bash
rsync -av /data /backup
if [ $? -eq 0 ]; then
  apprise -t "Backup Complete" -b "Daily backup finished successfully"
    "signal://localhost:9922/+306970000xyz"
else
  apprise -t "Backup FAILED" -b "Daily backup encountered errors"
    "signal://localhost:9922/+306970000xyz"
fi

System Health Checks

# Check disk space
USAGE=$(df / | tail -1 | awk '{print $5}' | sed 's/%//')
if [ $USAGE -gt 80 ]; then
  apprise -t "Disk Warning" -b "Root partition is ${USAGE}% full"
    "signal://localhost:9922/+306970000xyz"
fi

Advanced: Adding More Notification Channels

The beauty of Apprise is that adding new notification targets is as simple as adding a new URL:

# Add Slack
curl -X POST http://localhost:8800/add/slack
  -d "urls=slack://workspace/token"

# Add Email (Gmail)
curl -X POST http://localhost:8800/add/email
  -d "urls=mailto://user:password@gmail.com"

# Add ntfy (push notifications to mobile)
curl -X POST http://localhost:8800/add/ntfy
  -d "urls=ntfy://topic"

# Add Discord
curl -X POST http://localhost:8800/add/discord
  -d "urls=discord://webhook-id/webhook-token"

Troubleshooting

Signal Messages Not Sending

1. Check if Signal CLI is working:

   curl http://localhost:9922/v1/about

2. Verify your number is registered:

   curl http://localhost:9922/v1/send +306970000xyz -d "message=test"

3. Check container logs:

   docker logs signal-api

That’s it !

-Evaggelos Balaskas

LiteLLM AI Gateway (LLM Proxy)

This project shows a simple pattern: run multiple local model servers, place LiteLLM in front of them, and expose one OpenAI-compatible endpoint for apps to use.

In this setup, Lite LLM sits on http://127.0.0.1:4000/v1 and routes requests to:

• LM Studio on port 1234
• vLLM on port 8000
• Osaurus on port 1337

That gives you one clean API for local testing, model switching, and app integration.

Why this setup is useful

If you already have local models running in different tools, LiteLLM gives you one gateway instead of several different endpoints.

That means you can:

• keep one API base URL
• switch models by alias
• expose multiple local backends behind the same interface
• plug the endpoint into apps that expect an OpenAI-style API

What this article shows

This guide walks through the full flow:

1. LiteLLM exposes /v1/models and returns the three configured aliases.
2. A chat app can select one of those LiteLLM model names and answer normally.
3. Khoj can be pointed at LiteLLM by setting the API base to http://127.0.0.1:4000/v1.
4. Khoj chat models can then use a LiteLLM alias such as ministral-lmstudio.

Quick start

Use Python 3.12 or 3.13 for the LiteLLM virtual environment to avoid uvloop path a known incompatibility with Python 3.14’s asyncio internals.

Set up a virtual environment

virtualenv -p python3.12 venv-litellm/
cd venv-litellm/
source ./bin/activate

pip install "litellm[proxy]"

LiteLLM Configuration

Create config.yaml:

Below is an example based on my local setup.

model_list:
  - model_name: ministral-lmstudio
    litellm_params:
      model: openai/mistralai/ministral-3-3b
      api_base: http://127.0.0.1:1234/v1
      api_key: lmstudio

  - model_name: llama3-vllm
    litellm_params:
      model: openai/mlx-community/Llama-3.2-3B-Instruct-4bit
      api_base: http://127.0.0.1:8000/v1
      api_key: vllm

  - model_name: qwen3-osaurus
    litellm_params:
      model: openai/qwen3.5-0.8b-mlx-4bit
      api_base: http://127.0.0.1:1337/v1
      api_key: osaurus

You need the openai/ prefix for each model to use LiteLLM’s OpenAI-compatible provider.

Start LiteLLM

litellm --config config.yaml --port 4000

If LiteLLM starts correctly, you should see the three model aliases loaded.

Test the proxy

List models:

curl -s http://127.0.0.1:4000/v1/models | jq .

Test chat

Send a chat request:

curl -s http://127.0.0.1:4000/v1/chat/completions
  -H "Content-Type: application/json"
  -d '{
    "model": "ministral-lmstudio",
    "messages": [
      {"role": "user", "content": "Capital of Greece?"}
    ]
  }' | jq .

If everything is wired correctly, LiteLLM will forward the request to the matching backend and return a normal OpenAI-style response.

A typical result looks like this:

{
  "id": "chatcmpl-oukopuooxti6xg92g6qx9b",
  "created": 1773142359,
  "model": "ministral-lmstudio",
  "object": "chat.completion",
  "system_fingerprint": "mistralai/ministral-3-3b",
  "choices": [
    {
      "finish_reason": "stop",
      "index": 0,
      "message": {
        "content": "The capital of Greece is **Athens**.",
        "role": "assistant",
        "provider_specific_fields": {
          "refusal": null
        }
      },
      "provider_specific_fields": {}
    }
  ],
  "usage": {
    "completion_tokens": 10,
    "prompt_tokens": 539,
    "total_tokens": 549
  },
  "stats": {}
}

Connect LiteLLM to Khoj

You can use the same LiteLLM endpoint with any app that supports an OpenAI-style API. In this example, I use Khoj.

Once LiteLLM is running, Khoj only needs one API configuration:

• Name: litellm
• API key: litellm
• API base URL: http://127.0.0.1:4000/v1

Then create a chat model in Khoj using one of the LiteLLM aliases, for example:

• ministral-lmstudio

That is the key idea of this project: Khoj does not need to know whether the model is coming from LM Studio, vLLM, or Osaurus. It only talks to LiteLLM.

Khoj AI model API configuration

Khoj chat model configuration

Khoj using a LiteLLM chat model

That’s it!

If you want to use Claude Code together with Osaurus, there are two different pieces to understand:

1. Model backend — the LLM that answers your prompts
2. MCP tools — the tools Claude Code can call

This is the most important idea:

• Osaurus MCP gives Claude Code access to tools
• Osaurus API can also be used as the model backend, if your setup supports it

These are separate.

Install Claude Code and Osaurus

Let’s start by installing both tools via homebrew on a macbook.

Disclaimer: I like asaurus because it’s small and amazing, I find Ollama big and ugly in macbook.

claude code installation

brew install --cask claude-code

osaurus

brew install --cask osaurus

Open osaurus ui to setup osaurus, in this blog post we will not cover this.

language models

At some point you will download a couple LLMs or SLMs to start with osaurus and you should already have install some tools.

curl -s http://localhost:1337/v1/models | jq .

{
  "data": [
    {
      "id": "llama-3.2-3b-instruct-4bit",
      "created": 1772877371,
      "object": "model",
      "owned_by": "osaurus",
      "root": "llama-3.2-3b-instruct-4bit"
    },
    {
      "id": "qwen3-vl-4b-instruct-8bit",
      "created": 1772877371,
      "object": "model",
      "owned_by": "osaurus",
      "root": "qwen3-vl-4b-instruct-8bit"
    },
    {
      "id": "qwen3.5-0.8b-mlx-4bit",
      "created": 1772877371,
      "object": "model",
      "owned_by": "osaurus",
      "root": "qwen3.5-0.8b-mlx-4bit"
    }
  ],
  "object": "list"
}

status

❯ osaurus status
running (port 1337)

tools

❯ osaurus tools list
osaurus.browser  version=1.2.0
osaurus.fetch  version=1.0.2
osaurus.filesystem  version=1.0.3
osaurus.git  version=1.0.3
osaurus.images  version=1.0.3
osaurus.macos-use  version=1.2.1
osaurus.search  version=1.0.4
osaurus.time  version=1.0.3
osaurus.vision  version=1.0.1

Connect Claude Code to Osaurus via a MCP server

So by default claude code with autostart an interactive configuration setup to connect with your anthropic subscription or with any major ai subscription. We want to override this behaviour to enable claude to connect with osaurus. best way to do that is via an mcp server.

Create ~/.claude.json:

cat > ~/.claude.json <<EOF
{
  "theme": "dark-daltonized",
  "hasCompletedOnboarding": true,
  "mcpServers": {
    "osaurus": {
      "command": "osaurus",
      "args": [
        "mcp"
      ]
    }
  }
}
EOF

This tells Claude Code to start Osaurus as an MCP server.

Note on hasCompletedOnboarding: Setting this to true prevents a startup error where Claude Code tries to connect to Anthropic’s servers before your local endpoint is configured. It is not required for the MCP setup itself, but it avoids a confusing first-run failure.

Note on MCP config location: MCP servers must be defined in ~/.claude.json (or a project-local .mcp.json). Placing them in ~/.claude/settings.json will not work — that file is for environment variables and permissions, not MCP server definitions.

Configure Claude Code to use Osaurus as the model endpoint

Create ~/.claude/settings.json:

mkdir -p ~/.claude/

cat > ~/.claude/settings.json <<EOF
{
  "env": {
    "ANTHROPIC_BASE_URL": "http://127.0.0.1:1337",
    "ANTHROPIC_AUTH_TOKEN": "osaurus",
    "ANTHROPIC_MODEL": "qwen3-vl-4b-instruct-8bit"
  }
}
EOF

This does three things:

• points Claude Code to your local Osaurus server
• authenticates with the local Osaurus endpoint using a static token
• selects the model to use

Note on ANTHROPIC_MODEL vs ANTHROPIC_DEFAULT_SONNET_MODEL: ANTHROPIC_MODEL sets the model directly and is the simpler choice when Osaurus exposes a single model. ANTHROPIC_DEFAULT_SONNET_MODEL overrides only the model Claude Code uses when it internally requests a “sonnet”-class model — useful if you want different models for different internal roles, but unnecessary for a basic local setup.

and

Claude Code requires custom auth token values to be explicitly approved. ANTHROPIC_AUTH_TOKEN is for that

Without this, Claude Code may still prompt for authentication even though your token is set.

Start Claude Code

Run:

claude

Inside Claude Code, you can check your setup with:

/status

Simple mental model

Think of it like this:

• Model = the brain
• MCP = the toolbox

Changing the model does not remove the tools.

That is enough to get started.

Brave’s built-in privacy-first AI assistant, Leo, supports connecting to a local OpenAI-compatible server. This means your conversations never leave your machine — no cloud, no telemetry, just your browser talking to your own model.

This guide uses Osaurus on a MacBook M4 Pro, running the qwen3.5-0.8b-mlx-4bit model as a local example. Any OpenAI-compatible local server (LM Studio, Ollama, llama.cpp, etc.) will work the same way.

About the Model

Qwen3.5-0.8B is Alibaba’s latest small language model, released in March 2026. Despite its compact size, it is a native multimodal model — meaning it supports both text and vision (image understanding) out of the box. It runs efficiently on Apple Silicon via MLX quantization, making it an excellent fit for local inference on a MacBook M4 Pro with minimal RAM usage.

The mlx-4bit suffix means the model weights are 4-bit quantized for Apple Silicon using the MLX framework — fast, low-memory, and runs entirely on-device.

Prerequisites

• Brave Browser installed (check latest version)
• A local LLM server running and reachable at http://localhost:<port>
• Your server responds to POST /v1/chat/completions (OpenAI-compatible API)

Verify your server is working before continuing:

curl -s -X POST http://localhost:1337/v1/chat/completions
  -H "Content-Type: application/json"
  -d '{
    "model": "qwen3.5-0.8b-mlx-4bit",
    "messages": [{"role": "user", "content": "Say hello"}]
  }' | jq .

You should get a JSON response with a choices[0].message.content field. If that works, you’re ready.

example output

{
  "id": "chatcmpl-88053214C2DC",
  "object": "chat.completion",
  "created": 1772783955,
  "model": "qwen3.5-0.8b-mlx-4bit",
  "choices": [
    {
      "finish_reason": "stop",
      "message": {
        "content": "Hello! How can I help you today? 😊",
        "role": "assistant"
      },
      "index": 0
    }
  ],
  "usage": {
    "prompt_tokens": 2,
    "completion_tokens": 8,
    "total_tokens": 10
  }
}

Step 1 — Enable Required Brave Flags

Before Leo can connect to a local server, you need to enable two feature flags in Brave.

Open a new tab and go to:

brave://flags

Search for and enable each of the following:

After enabling both flags, click Relaunch to restart Brave.

Step 2 — Open Leo Settings

Once Brave restarts, open Leo settings by navigating to:

brave://settings/leo-ai

Or open the Leo sidebar (chat bubble icon) → click the Settings gear icon.

Step 3 — Add a Custom Model

In the Leo settings page, scroll down to Bring your own model and click Add new model.

Fill in the fields as follows:

Click Save model.

Note: The server endpoint must be the full path including /v1/chat/completions, not just the base URL.

Step 4 — Select Your Local Model

Back in the Leo chat panel:

1. Click the model selector dropdown (shows the currently active model name).
2. Select the model you just added — e.g. Osaurus.

Leo will now route all requests to your local server.

Step 5 — Start Chatting

Type a message in the Leo input box and press Enter.

How It Works

Your request goes to http://localhost:1337/v1/chat/completions — entirely on your machine. Nothing is sent to Brave’s servers or any external service.

You type in Leo
      │
      ▼
Brave sends POST /v1/chat/completions
      │
      ▼
localhost:1337  (your local server — Osaurus)
      │
      ▼
Model inference on Apple Silicon (MLX / 4-bit quantized)
      │
      ▼
Response streams back to Leo in your browser

No internet required after setup. No data leaves your device.

Tips

• Model name must match exactly what your server reports — check it with:

curl http://localhost:1337/v1/models | jq .

eg.

{
  "data": [
    {
      "object": "model",
      "id": "llama-3.2-3b-instruct-4bit",
      "created": 1772791159,
      "root": "llama-3.2-3b-instruct-4bit",
      "owned_by": "osaurus"
    },
    {
      "object": "model",
      "id": "qwen3.5-0.8b-mlx-4bit",
      "created": 1772791159,
      "root": "qwen3.5-0.8b-mlx-4bit",
      "owned_by": "osaurus"
    }
  ],
  "object": "list"
}

• Leo context features (summarize page, ask about selected text) also work with local models — Leo includes the page content as part of the prompt automatically.
• Since Qwen3.5-0.8B supports vision, with Vision Support enabled you can paste or drag images into Leo and the model will analyze them — all locally.
• Start your local server before opening Brave, or you’ll get a connection error when Leo tries to reach it.

That’s it. You now have a fully local, private AI assistant inside your browser — no accounts, no subscriptions, no data leaving your machine.

Want to run modern LLMs locally — with an OpenAI-compatible API, multimodal support, and strong performance on Apple Silicon? This beginner-friendly guide walks you through everything from installation to your first inference request.

No prior ML experience required.

What is vllm-mlx?

vllm-mlx is a community-driven inference server built specifically for Apple Silicon Macs. It uses MLX, Apple’s machine learning framework designed for M-series chips, and exposes an OpenAI-compatible HTTP API so you can drop it in wherever you’d use the OpenAI SDK.

Think of it as a full, self-contained AI server stack that runs entirely on your Mac.

How does it differ from official vLLM?

Important: vllm-mlx is not a plugin or fork of official vLLM. It’s a separate framework built from the ground up for Macs.

Architecture overview

When you run vllm-mlx serve, you get a layered system:

Why use vllm-mlx?

It’s the right tool if you want:

• A full-featured local AI server on Apple Silicon
• Text and multimodal inference in a single server
• OpenAI-compatible APIs out of the box
• Fully offline inference — no cloud, no data leaving your machine

System requirements

• macOS with Apple Silicon (M1/M2/M3/M4)
• Python 3.10+
• 16 GB RAM minimum recommended (larger models require more)

Step 1 — Create a clean Python environment

Never install ML tooling into your global Python. Use an isolated virtual environment:

python3 -m venv ~/.venv-vllm-mlx
source ~/.venv-vllm-mlx/bin/activate

Once activated, your shell prompt should change to something like:

(venv-vllm-mlx) yourname@macbook %

Alternatively, with virtualenv:

virtualenv venv-vllm-mlx
cd venv-vllm-mlx
source ./bin/activate

Step 2 — Install vllm-mlx

pip install vllm-mlx

Verify the installation:

pip list | grep vllm

You should see vllm-mlx in the output.

Step 3 — Start your first model server

We’ll use a 4-bit quantized Llama 3.2 model — small, fast, and a good starting point.

vllm-mlx serve mlx-community/Llama-3.2-3B-Instruct-4bit --port 8010

This command will:

1. Download the model from HuggingFace (first run only)
2. Load it into the MLX backend
3. Start an HTTP API server on port 8010

You’ll see log output showing the model loading and the server starting on 0.0.0.0:8010.

Step 4 — Verify the server

Health check

curl -s http://localhost:8010/health | jq .

Expected output:

{
  "status": "healthy",
  "model_loaded": true,
  "model_name": "mlx-community/Llama-3.2-3B-Instruct-4bit",
  "model_type": "llm",
  "engine_type": "simple",
  "mcp": null
}

List available models

curl -s http://localhost:8010/v1/models | jq .

Expected output:

{
  "object": "list",
  "data": [
    {
      "id": "mlx-community/Llama-3.2-3B-Instruct-4bit",
      "object": "model",
      "created": 1772701579,
      "owned_by": "vllm-mlx"
    }
  ]
}

Step 5 — Send a chat request

Use the OpenAI-compatible /v1/chat/completions endpoint:

curl -s http://127.0.0.1:8010/v1/chat/completions
  -H "Content-Type: application/json"
  -d '{
    "model": "mlx-community/Llama-3.2-3B-Instruct-4bit",
    "messages": [
      {"role": "user", "content": "Hello! What is the capital of Greece?"}
    ],
    "max_tokens": 100
  }' | jq .

Expected response:

{
  "id": "...",
  "object": "chat.completion",
  "choices": [
    {
      "message": {
        "role": "assistant",
        "content": "The capital of Greece is Athens."
      }
    }
  ]
}

You’re now running a local LLM server on your Mac.

Running larger models (advanced)

For high-memory Macs (64 GB+ recommended), you can run much larger models with additional flags:

vllm-mlx serve Qwen/Qwen3.5-35B-A3B-GPTQ-Int4
  --port 8010
  --max-tokens 262144
  --reasoning-parser qwen3

What you can do next

With your local server running, you can connect it to the broader AI tooling ecosystem by pointing any OpenAI-compatible client at http://localhost:8010/v1:

• Open WebUI — browser-based chat UI
• LangChain or LlamaIndex — agent and RAG pipelines
• OpenAI Python SDK — just set base_url="http://localhost:8010/v1"
• Embeddings and multimodal models — swap in a different model and the same API applies

Why Apple Silicon works so well here

On a discrete GPU setup (NVIDIA), model weights must be copied over PCIe from system RAM to VRAM before inference can begin. Apple Silicon eliminates this bottleneck entirely — the CPU and GPU share the same unified memory pool. Combined with Apple’s high memory bandwidth, this makes MLX extremely efficient for inference on models that fit in RAM.

Multimodal routing

When using a vision or audio model, the server adds an extra routing step:

Image / Audio input
        ↓
Multimodal Router (mlx-vlm / audio pipeline)
        ↓
LLM reasoning
        ↓
Text output

No additional services are required — it’s built into the same server process.

How vllm-mlx differs from official vLLM under the hood

Official vLLM:   App → vLLM Engine → CUDA kernels → NVIDIA GPU
vllm-mlx:        App → vllm-mlx Server → MLX tensors → Apple GPU

These are entirely different acceleration stacks. vllm-mlx doesn’t use or depend on any CUDA code.

That’s it. A local, fully offline, OpenAI-compatible LLM server running natively on your Mac.