Ghostwheel is a coding agent that runs locally on your Windows computer with unlimited usage.
Raw AI agent power, running locally on your hardware.
Ghostwheel is a Windows desktop application that integrates a local developer agent with on-device LLM inference. Built on the Merlin Framework, it packages a thin Tauri runtime controlling a pinned native llama-server sidecar. By running prompt context, file edits, and terminal executions entirely inside your local hardware boundary, Ghostwheel eliminates API token billing, network lag, and prompt telemetry.
Note: This repository is the public release tracker, documentation home, and community hub for Ghostwheel. The core runtime engine is proprietary and closed-source.
To run Ghostwheel's local inference engine (llama-server) with active hardware acceleration, your machine must meet the following specifications:
| Hardware Component | Minimum Requirement | Recommended Specification |
|---|---|---|
| Processor (CPU) | Intel Core i5 / AMD Ryzen 5 | Intel Core i7 / AMD Ryzen 7 (or newer) |
| System Memory (RAM) | 16 GB DDR3 or faster | 32 GB DDR4 / DDR5 |
| Dedicated Graphics (GPU) | 4 GB VRAM (NVIDIA CUDA or Vulkan) | 8 GB+ Dedicated VRAM (NVIDIA RTX) |
| Storage Space | 8 GB free space (SSD recommended) | 16 GB free space (NVMe SSD) |
Cloud-based developer agents are powerful, but they expose your proprietary code to third-party servers, require complex API token subscriptions, and fail completely when offline. Ghostwheel provides a local-first alternative without sacrificing capability.
| Feature | Ghostwheel (Local) | Cloud Agents (SaaS) | Raw CLI Wrappers |
|---|---|---|---|
| Data Privacy | 🔒 100% Local (No telemetry) | 🔒 Local | |
| Inference Cost | 💰 Free / Unlimited (Your hardware) | 💸 Per-token billing | 💸 API subscription keys |
| Offline Capability | ❌ Requires internet | ❌ Requires internet | |
| Hardware Acceleration | ⚡ CUDA & Vulkan auto-selection | ❌ Runs in cloud VM | |
| Sandboxed Tooling | 🛡️ SSRF & Command Confirmations | ❌ Runs naked on host |
On startup, Ghostwheel executes a full diagnostic checks sequence to verify dependencies and hardware allocation:
$ ghostwheel boot
[ok] webview2 runtime: 121.0.x
[ok] llama-server: b9729 (cuda-13.3-x64)
[ok] model: Merlin-9B-Coder-Q5_K_M.gguf
sha256: 7e3c…verified
[ok] sqlite: %LOCALAPPDATA%\ProjectMerlin\merlin.db
[ok] license: active · grace 14d
[ok] network egress: 0 bytes
$ ready_Ghostwheel is distributed as a signed MSI installer for Windows 10 and 11.
- Download: Fetch the installer matching your hardware profile from our website's Download Center (routed at
#releases). - Verify Checksum: Always verify the installer's integrity by comparing the SHA-256 hash in PowerShell:
Get-FileHash .\Ghostwheel-Installer-CUDA.msi -Algorithm SHA256
- Run: Run the installer and input your waitlist license key on first boot.
For advanced configuration (VRAM allocation, context scaling, custom model manifests), visit our website's Docs section (routed at #docs).
Ghostwheel is currently in a private pre-alpha phase, with active builds compiling on our local loop.
- To lock in your spot and receive early-access release keys, join the waitlist at our website's Waitlist Form (routed at
#download). - If you want to speed up the compiler and back Rydell's development, you can Buy Rydell a Coffee via the link on our homepage or backer page (routed at
#backer).
While Ghostwheel is built on top of open-source libraries (such as llama.cpp and Tauri), the core orchestration layers, safety guards, and commercial integrations are proprietary.
This model allows us to:
- Guarantee rigorous verification bounds on local command-execution safety before wide release.
- Fund full-time development and active compiler maintenance.
- Ensure absolute compliance with licensing and distribution policies.
We are fully committed to open-source compliance; all third-party notices and licenses are documented on the website's Third-Party Notices page (routed at #licenses).
- Bug Reports & Feature Requests: Please use our Issue Tracker to report bugs or suggest enhancements.
- Security Disclosures: Found a vulnerability? Please refer to our Security Policy for secure reporting channels.
- General Support: File a support ticket at our website's Support Ticket Form (routed at
#support).
