DeepHorus is an innovative tech media and intelligence platform focused on AI, blockchain, startups, and emerging technologies.
Published on

Riffn. An instant voice link with your AI agents and local models

Authors
  • avatar
    Name
    DeepHorus
    Twitter

Riffn, a new iOS app, lets developers talk to AI coding agents running on their own computers using just an iPhone. The app pairs with a local coding agent over a Tailscale network, allowing sessions to continue in the background with the phone locked. It is designed for hands-free use while driving, walking, or doing chores.

The setup involves three steps: running a small bridge in the repository, pairing via QR code, and starting a session. The bridge, @riffn/bridge, is an open-source Node helper around 100 KB with zero runtime dependencies and Apache-2.0 license. It binds to the tailnet by default and refuses public internet exposure unless deliberately overridden.

Riffn works with Claude Code, which must already be installed and authenticated on the computer. It does not include a model subscription or API credits; instead, it uses the existing CLI and provider account. Users can run other models like OpenAI or GLM through the Claude Code harness via gateways such as LiteLLM.

One key feature is voice-based switching between agents or models. Users can assign one-word names and say commands like "Switch to Gloria" to move mid-session without touching the phone. Each bridge gets its own port and project state, enabling multiple agents to run side by side on one machine.

The developer made deliberate security choices, including disabling direct Codex bridging on Windows due to potential file-read bypass. Routing through the Claude Code harness keeps safety guarantees enforced on the user's machine. Users should note that Riffn uses OpenAI API billing, and LiteLLM is third-party software that holds the API key.

Riffn addresses a growing need for hands-free interaction with AI tools in everyday contexts. By keeping sessions active in the background and supporting earbuds and CarPlay, it aims to make coding assistance as accessible as a phone call. Future updates may expand supported agents and models, but for now, it offers a novel way to stay connected to AI agents on the go.