Control Center
See gateway health, CPU/RAM pressure, tokens used and OpenClaw version before anything breaks.
local-first
your machine
cloud optional
no signup
Native Control Center for OpenClaw
Install LocalClaw once. Control your local models, agents, channels and scheduled OpenClaw work from a native macOS dashboard.
Installer $49 or free model finder. No signup. No prompts collected.
LocalClaw 1.0.175 · Developer workspace upgraded
LocalClaw v1.0.175
Cloud LLM
OAuth LLM
Local LLM
LocalClaw
Version 1.0.175 (build 326)
⌁
Online
Gateway
google/gemma-4-e4b
▤
Healthy
System
CPU 28% · RAM 31.2/32 GB
▥
299.2K
Tokens used
6 requests · 30 days
▣
1 active
Channels
22 available · 1 account
OpenClaw readiness
Ready
5/5 checks passing
●Gateway
●Chat
●Model
●Channel
●Skill
Next best actions
▱ Open OpenClaw Chat
Everything essential is ready.
Everything essential is ready.
System Load
CPU38%
RAM31.3 / 32.0 GB
Swap1.85 / 3.00 GB
0 MB
OpenClaw
OpenClaw
0 MB
LM Studio
LM Studio
0 MB
Node
Node
Recent activity
● Managed skills: /Users/redsun/.openclaw/skills
● Workspace: /Users/redsun/.openclaw/workspace
● Channel inventory refreshed.
Working now
▱ChatReady●
▣Channels1 connected · 22 available●
⌁Skills18 active · 58 installed●
▤Modelsgoogle/gemma-4-e4b●
▥Automation6 active · 7 jobs●
Operations
▸OpenClaw Chat›
▸Connect channels›
▸Manage models›
▸Add skills›
// REAL USE CASE
Kimi K3 built a playable 3D FPS inside LocalClaw
The model worked through OpenRouter while LocalClaw kept the project, AI activity, files and live preview in one native workspace. The result is Iron Descent, playable in the same screen where it was built.
- Model
- Kimi K3
- Workspace
- Developer
- Feedback
- Live preview
- Result
- Playable FPS
// PLAYABLE PUBLIC BUILD
The finished browser FPS is live, and its source is public for inspection.
- 01
Describe the change
Give the model a concrete product brief in the project chat.
- 02
See the work
Follow tool activity, elapsed time and the files being changed.
- 03
Test it live
Use the interactive preview without leaving LocalClaw.
- 04
Iterate in context
Keep the project state and refine the result instead of starting over.
Real workspace · visible AI activity · live preview · finished gameplay
// THE APP
Everything to run OpenClaw locally, in one place
A native macOS control center for the local AI stack you already own: models, agents, channels, schedules and system health.
Models
Switch between Local LLM, Cloud LLM and OAuth LLM without hand-editing config files.
Agents
Create autonomous OpenClaw agents, assign models and keep their local workspace visible.
Cron Jobs
Schedule recurring local tasks and see what will run, when it runs and which agent owns it.
Kanban
Turn local automation into visible work: backlog, ready, doing, review and done.
Chat · Skills · Channels
Use beta dev tools, channel connections and local skills from the same dashboard.
// LOCAL FIT ENGINE
Benchmarks are useful. Fit is what makes local AI work.
LocalClaw ranks models by practical local fit: your memory headroom, context target, use case, quantization and install path. Every recommendation explains why it was picked.
01
Hardware fit first
RAM, VRAM, OS and model size decide whether a model feels instant or painful.
Apple M4 · 16 GB → laptop-safe picks
02
Context has a cost
Long context increases KV cache memory, so LocalClaw warns before a model becomes slow.
32K context → extra headroom check
03
Use-case weighting
Coding, chat, reasoning, vision and speed priorities change the shortlist transparently.
Coding → code benchmark + model tags
04
Installability matters
The best model is the one you can actually load locally with the right quantization.
Q5_K_M / Q4_K_M shown upfront
// RECENTLY VERIFIED
Fresh local AI worth checking
New open-weight releases are screened for a real local install path before they enter LocalClaw.
Jul 2026 · Coding MoE
Laguna XS 2.1
33B total, 3B active, 262K context and official Q4_K_M GGUF for 36 GB machines.
Jul 2026 · 1-bit / ternary
Bonsai 27B
A 27B multimodal model compressed to 3.9GB or 7.2GB with dedicated local runtimes.
Jul 2026 · MoE
Hy3
295B total, 21B active, 256K context. A large-workstation local target.
// START HERE
Not sure what to open next?
Pick a practical path. Each guide connects hardware, RAM, model fit and the next LocalClaw action.
Hardware
Best local LLMs for Mac mini M4
Start from the machine and choose models that fit unified memory.
RAM
Best local LLMs for 16GB RAM
Laptop-safe picks that avoid memory pressure and slow inference.
Use case
Best local LLMs for coding
Private code assistants, agents, debugging and repo workflows.
Voice
Best local TTS for voice cloning
Speech models for private voice workflows and offline pipelines.
LocalClaw
Download it like an app
- ▸ Guided macOS setup for OpenClaw
- ▸ Managed local models and runtime checks
- ▸ Agents, channels, skills and schedules in one UI
- ▸ Local-first: local models and prompts stay on your machine
Manual setup
Hours in the terminal
- ▸ Install dependencies by hand
- ▸ Edit config files and CLI flags
- ▸ Track models, agents and channels separately
- ▸ Debug local runtime issues after the fact
// CATALOGUE
A small sample of what LocalClaw tracks
Frontier
DeepSeek V4 Pro
1.6T MoE · 49B active · local catalogue
Agentic
GLM-5.2
744B MoE · 1M context · workstation class
Laptop
Qwen 3.5 9B
Balanced local model for 16 GB+
New
Gemma 4 12B
Unified multimodal sweet spot for 16 GB+
Voice
NeuTTS Air
Real-time CPU TTS and voice cloning
New voice
Dots TTS MF
2B zero-shot voice cloning · Apache 2.0
Frequently Asked Questions
What is LM Studio?
LM Studio is a free desktop application that lets you run Large Language Models (LLMs) locally on your computer. No internet needed, no data sent anywhere. It provides a chat interface similar to ChatGPT but everything runs on YOUR hardware.
What is quantization (Q4, Q5, Q8)?
Quantization is a compression technique that reduces model size while preserving most of the quality. Think of it like JPEG compression for images. Q4 = more compressed (smaller, slightly lower quality), Q8 = less compressed (larger, nearly original quality). Q5_K_M is the sweet spot for most users.
How much RAM do I need to run a local AI model?
Rule of thumb: the model file size plus 2-3 GB for the system. A 5 GB model needs at least 8 GB RAM. On macOS with Apple Silicon, the unified memory makes things more efficient. On Windows/Linux with a GPU, VRAM helps offload the model.
Apple Silicon vs NVIDIA GPU for local AI?
Apple Silicon (M1-M4) uses unified memory, meaning your entire RAM is available for the model. This is incredibly efficient. NVIDIA GPUs are faster for inference but limited by VRAM (typically 8-24 GB). Both are great choices.
Is my data private when using LocalClaw?
Yes. The model finder runs in your browser and does not collect prompts, hardware specs, or selected models. When using LM Studio with recommended models, inference runs locally on your machine with no cloud API calls.
What are the best local AI models in 2026?
For 8 GB RAM: Qwen 3.5 4B or Gemma 4 E4B. For 16 GB: Gemma 4 12B, Qwen 3.5 9B, GLM 4.6 Air 12B or Mistral Small 3.2 24B (tight). For 32 GB+: Gemma 4 31B, Qwen 3 Next 80B/3B MoE or Qwen 3 Coder 30B. For reasoning: Kimi K2 Thinking, DeepSeek V3.2 Exp, or Hermes 4 70B. For coding: MiniMax M2 and Qwen 3 Coder. For vision: Qwen 3 VL 32B or Gemma 4 multimodal.
What is OpenClaw?
OpenClaw is an open-source AI gateway for agents, channels, scheduled work and model providers. Connect it to LM Studio or Ollama for local inference, or choose an optional cloud provider. Local models can keep prompts on your machine; cloud providers and connected online services follow their own data and API policies.
What is free and what is paid?
The LocalClaw web model finder is free: use it to choose the right LLM/TTS model for your hardware. The LocalClaw beta app is the optional native macOS app that simplifies setup, handles activation, supports updates and helps manage your local AI stack. No terminal-first workflow. The current beta lifetime offer is $49 one-time, no subscription, no recurring fees. Your license is valid forever. See pricing →