Instructions to use badtheorylabs/Macaw-4bit-MLX with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use badtheorylabs/Macaw-4bit-MLX with MLX:

# Make sure mlx-lm is installed
# pip install --upgrade mlx-lm

# Generate text with mlx-lm
from mlx_lm import load, generate

model, tokenizer = load("badtheorylabs/Macaw-4bit-MLX")

prompt = "Write a story about Einstein"
messages = [{"role": "user", "content": prompt}]
prompt = tokenizer.apply_chat_template(
    messages, add_generation_prompt=True
)

text = generate(model, tokenizer, prompt=prompt, verbose=True)

Notebooks
Google Colab
Kaggle
Local Apps Settings
LM Studio

How to use badtheorylabs/Macaw-4bit-MLX with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "badtheorylabs/Macaw-4bit-MLX"

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "mlx-lm": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "badtheorylabs/Macaw-4bit-MLX"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

OpenClaw new

How to use badtheorylabs/Macaw-4bit-MLX with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "badtheorylabs/Macaw-4bit-MLX"

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "badtheorylabs/Macaw-4bit-MLX" \
  --custom-provider-id mlx-lm \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

MLX LM

How to use badtheorylabs/Macaw-4bit-MLX with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Interactive chat REPL
mlx_lm.chat --model "badtheorylabs/Macaw-4bit-MLX"

Run an OpenAI-compatible server

# Install MLX LM
uv tool install mlx-lm
# Start the server
mlx_lm.server --model "badtheorylabs/Macaw-4bit-MLX"
# Calling the OpenAI-compatible server with curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
   -H "Content-Type: application/json" \
   --data '{
     "model": "badtheorylabs/Macaw-4bit-MLX",
     "messages": [
       {"role": "user", "content": "Hello"}
     ]
   }'

Hermes Agent

How to use badtheorylabs/Macaw-4bit-MLX with Hermes Agent:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "badtheorylabs/Macaw-4bit-MLX"

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default badtheorylabs/Macaw-4bit-MLX

Run Hermes

hermes

🦜 Macaw · 4-bit MLX

The build that runs on your Mac. 1.5 GB, ~2 GB of memory, about a second per request on an M2.

Ask for something in plain English and it does it — sends the email, finds the file, reads the PDF, tells you what's draining your battery. Nothing leaves the machine.

👉 badtheorylabs.com/macaw

Install

pip install mlx-lm

mlx_lm.generate --model badtheorylabs/Macaw-4bit-MLX \
  --prompt "what's my battery at?"

Or serve it locally:

mlx_lm.server --model badtheorylabs/Macaw-4bit-MLX --port 8138

Then talk to it like any OpenAI-compatible endpoint, passing your tools:

import requests

tools = [{"type": "function", "function": {
    "name": "battery_status",
    "description": "Report battery percentage and charging state."}}]

r = requests.post("http://127.0.0.1:8138/v1/chat/completions", json={
    "model": "badtheorylabs/Macaw-4bit-MLX",
    "messages": [{"role": "user", "content": "what's my battery at?"}],
    "tools": tools, "temperature": 0.0, "max_tokens": 128})

print(r.json()["choices"][0]["message"]["content"])
# <|tool_call_start|>[battery_status()]<|tool_call_end|>

You run the tool, feed the result back, and it answers in plain language. The full app — 97 macOS tools, parsing, and a confirmation gate on anything destructive — is at github.com/Badtheorylabs/Macaw.

What it's for

"email ada the q3 numbers and tell her i approved"
"why is my mac slow?"
"read ~/Documents/contract.pdf and summarise it"
"take a screenshot, make a folder called Shots, and move it there"
"am i free tomorrow afternoon?"

Mail · Calendar · Files · Notes · Reminders · Music · Safari · Chrome · System settings · Screen reading · Documents · Diagnostics

Specs


Download	1.5 GB
Memory	~2 GB running
Speed	~1.2 s per request (M2) · ~40 tok/s decode
Context	128K
Quantization	4-bit affine, group size 64
Requires	Apple Silicon (M1+), macOS 14+

Need BF16 for fine-tuning or GPU serving? That's badtheorylabs/Macaw.

Private by construction

There is no server. The model is on your disk, the tools run through macOS APIs on your machine, and nothing is transmitted. No account, no telemetry, no exceptions — including from us.

License

Derived from LFM2.5-2.6B under the LFM Open License v1.0. Free commercially below $10M annual revenue. App and tooling are MIT.

Downloads last month: -

Safetensors

Model size

0.4B params

Tensor type

BF16

U32

MLX

Hardware compatibility

4-bit

Model tree for badtheorylabs/Macaw-4bit-MLX

Base model

LiquidAI/LFM2.5-2.6B-Base

Finetuned

LiquidAI/LFM2.5-2.6B

Finetuned

badtheorylabs/Macaw

Quantized

(3)

this model