AirClaw ⚡
Run OpenClaw on ANY GPU with zero API cost
Stop paying $50–$150/month in API bills.
One command → local model on your machine → $0/month. Forever.
Supports RabbitLLM (newer models: Qwen2.5, DeepSeek, Phi-3) and AirLLM as fallback.
Runs 70B models on a 4GB GPU. Works on CPU too.
Install
pip install airclaw
Or from GitHub:
pip install git+https://github.com/nickzsche21/airclaw.git
Quick Start
# Full automated setup (do this first — one time only)
airclaw install
# Then in terminal 1:
airclaw start
# Then in terminal 2:
openclaw restart
Done. Your OpenClaw runs locally. Zero API cost.
Commands
| Command | What it does |
|---|---|
airclaw install |
Full automated setup |
airclaw start |
Start local LLM server (Mistral 7B default) |
airclaw start --model qwen |
Start with Qwen2.5 |
airclaw start --model deepseek |
Start with DeepSeek |
airclaw patch |
Auto-patch OpenClaw config |
airclaw patch --config ~/path/config.json |
Patch specific config |
airclaw restore |
Restore original config |
airclaw status |
Check if server is running |
Models
| Flag | Model | VRAM | Speed |
|---|---|---|---|
7b ← default |
Mistral-7B-Instruct | 4GB | ⚡⚡⚡ |
8b |
Llama-3-8B-Instruct | 6GB | ⚡⚡⚡ |
qwen |
Qwen2.5-7B | 4GB | ⚡⚡⚡ |
deepseek |
DeepSeek-7B | 4GB | ⚡⚡⚡ |
phi |
Phi-3-mini | 4GB | ⚡⚡⚡ fastest |
13b |
Llama-2-13B | 8GB | ⚡⚡ |
70b |
Llama-2-70B | 4GB | ⚡ slow |
Or pass any HuggingFace model ID:
airclaw start --model mistralai/Mixtral-8x7B-Instruct-v0.1
Backends
AirClaw automatically uses the best available backend:
- RabbitLLM (preferred) — newer models, faster, 4bit compression
- AirLLM (fallback) — battle-tested, wide compatibility
Install both:
pip install airclaw[all]
Or just one:
pip install airclaw[rabbitllm]
pip install airclaw[airllm]
How It Works
WhatsApp / Telegram / Discord
↓
OpenClaw
↓
AirClaw (localhost:4096) ← replaces OpenAI/Claude API
↓
RabbitLLM or AirLLM
↓
Local model on your GPU/CPU
AirClaw runs an OpenAI-compatible server on localhost. OpenClaw connects to it exactly like OpenAI — but everything runs on your machine.
Cost
| Setup | Monthly |
|---|---|
| OpenClaw + GPT-4o | $50–$150/mo |
| OpenClaw + Claude | $30–$120/mo |
| OpenClaw + AirClaw | $0/mo |
Requirements
- Python 3.10+
- OpenClaw installed
- 4GB+ GPU (or CPU — slower)
- 8–15GB disk for model weights (downloaded once)
License
MIT — free forever.
Star ⭐ if this saved you money.
Metadata
Release files for airclaw 2.0.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| airclaw-2.0.0.tar.gz | 10.2 kB | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| airclaw-2.0.0-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 19.7 kB
Release files / airclaw-2.0.0.tar.gz
| Download URL | airclaw-2.0.0.tar.gz |
|---|---|
| Size | 10.2 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
16a20aa76857ce158dad6daab00be978ddbef87cd1205575d394ce532dce01fd
|
|
BLAKE2b-256 checksum How to use checksums |
bb90f59f4cdc5487e766fb7cc0264489517cceb2514cb383ce26d7fa4fc5520c
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.13.5
|
Release files / airclaw-2.0.0-py3-none-any.whl
| Download URL | airclaw-2.0.0-py3-none-any.whl |
|---|---|
| Size | 9.5 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
604d7d694b2722b340d3ba9655ab059367dac6a3569ac01201671d297f32e87c
|
|
BLAKE2b-256 checksum How to use checksums |
227f18cba8ee50f9ab022024c20736c4c682e18a4a5e852270d728e4cb9e0157
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.13.5
|