Tela, written by Paul Parks, is aiming at a very specific pain point: getting to TCP services like SSH, RDP, HTTP, or database ports when the machine you want is behind NAT, the client machine is locked down, and a conventional VPN is either too heavy or simply not allowed. The repo describes Tela as a remote-access fabric built around encrypted WireGuard tunnels relayed over WebSocket, with no TUN device and no admin privileges required on either end.
That implies that spinning up a Tela network has a relatively light administrative burden in terms of system permissions. In this, it'd fit in the same space as Tailscale (or Headscale as a cleaner analog), or Wireguard or ngrok or Cloudflare.
The access levels required makes it interesting: Tela is not "just another tunnel.” The project is explicitly built around outbound-only connectivity, userspace WireGuard via gVisor netstack, and a hub that relays ciphertext rather than terminating the tunnel. The current stack consists of a client (tela), an agent/daemon (telad), a hub (telahubd), and a desktop UI called TelaVisor. The design doc is unusually explicit about the separation between the core fabric and any future platform wrapped around it.
The practical pitch is easy to understand: connect to home lab systems, dev machines, production boxes, or customer endpoints without opening inbound ports or forcing a full-network VPN onto every device. (Thus the "light administrative burden." Systems access doesn't require elevation the same way many tunnels would.) The use-cases doc keeps coming back to the same advantages: no port forwarding, no dynamic DNS, no client-side admin access, and service-level exposure instead of broad network access.
The project also appears to be honest about where it is. The status document says the core pieces work at proof-of-concept level, including the client, agent, hub, userspace WireGuard transport, UDP relay, and direct P2P fallback, while some of the more formal protocol machinery - like multiplexed channels and full token design - still remains unfinished. The gVisor dependency is going to impact network performance measurably, but given the project's targeting, may not matter.
All of that makes Tela more interesting, not less: it reads like a serious system under active construction rather than a marketing shell around vapor. It's also implemented and designed using "modern tooling," using AI as a partner rather than a source, and the strength of the design suggests that the technology is being used well rather than as a replacement for understanding.