dofastted/vm2api

What it solves

This project addresses the high risk of account suspension and service disruption when using third-party reverse proxies to access AI services like Claude and ChatGPT. Instead of using fragile HTTP emulation, it provides a production-grade gateway that runs official AI client processes within isolated virtual environments, ensuring stable, long-term access.

How it works

It manages a fleet of isolated virtual machines (VMs) or containers, each running an official AI client binary (such as Claude Code). Every slot is assigned a unique network egress (SOCKS5 proxy) to prevent cross-account contamination. The system acts as a protocol gateway, accepting standard API requests (OpenAI/Anthropic formats), cleaning them, and relaying them to the official client processes while simulating realistic hardware fingerprints and telemetry. It also includes a multi-layer risk management system that filters requests before they reach the upstream provider.

Who it’s for

Developers and infrastructure operators who need to build reliable, high-concurrency AI applications on top of subscription-based AI services without triggering anti-abuse mechanisms or facing frequent account bans.

Highlights

• Official Client Integration: Runs authentic official binaries rather than reverse-engineered HTTP mocks.
• Network Isolation: Enforces a strict one-VM-to-one-IP policy to prevent shared-IP bans.
• Hardware Emulation: Spoofs physical hardware fingerprints (MAC, SMBIOS, Machine-ID) to mimic real developer workstations.
• Risk Management: Features a multi-stage pre-hop gate system (distillation detection, regex filtering, and safety model scoring) to prevent policy violations.
• Cluster Management: Supports managing distributed VPS nodes via SSH with a centralized dashboard for monitoring quotas and costs.