Infercat Enables Secure P2P Sharing of Local AI Models via Invite Codes
A developer introduced Infercat, an open-source gateway and tunneling tool that lets users share access to their locally hosted AI models via end-to-end encrypted peer-to-peer tunnels using simple invite codes. It automatically integrates with popular local inference engines like Ollama, llama.cpp, vLLM, and LM Studio. Running local AI models often limits usage to home networks unless users navigate complex VPNs or port forwarding, driving them back to commercial cloud services. Infercat solves this pain point by turning personal hardware into private, manageable mini-clouds accessible from mobile devices or by friends. Built using Tailscale's tailcat library for P2P networking, Infercat exposes an OpenAI-compatible API supporting `/v1/chat/completions`, `/v1/models`, and `/v1/embeddings` while offering access control features like token rate limits, model allowlists, and key revocation. Web browser clients can chat directly without installing any software, while a local CLI command routes traffic locally for tools like Open WebUI, Cursor, or Claude Code.
## BACKGROUND
An AI inference engine is a software component that runs pre-trained machine learning models locally on hardware to process prompts and generate predictions. Peer-to-peer (P2P) tunneling technology creates secure, encrypted network connections directly between devices over the internet without requiring public IP addresses or traditional central VPN servers.