Multi-voice AI orchestration
One chat.
Many voices.
What Polyphon does
Polyphon is a free, local-first desktop app for orchestrating conversations between multiple AI voices simultaneously. You send a message — every voice in the session responds, reading each other's replies and building on what came before. A critic and a builder. A generalist and a specialist. A cloud model and a local one. Running together, in the same conversation.
Polyphon supports API voices (Anthropic Claude, OpenAI GPT, Google Gemini), CLI voices (Claude Code, Codex, GitHub Copilot), and any OpenAI-compatible endpoint — Ollama, LM Studio, vLLM, or your own. Mix and match freely. The question Polyphon is designed around is "What voices do I want in this room?" — not "Which single model should I use?"
Sessions can be broadcast (everyone responds) or directed (you @mention a specific voice). Compositions let you save and reuse your favorite configurations. Voices can use tools — reading and writing files, running commands, fetching URLs — scoped to a working directory you control.
How it's built
Polyphon is local-first. Your sessions, compositions, and settings live on your machine, encrypted at rest with AES-256 via SQLCipher. Nothing is routed through a backend. There is no account. There is no telemetry — ever.
API keys are read from your shell environment and never leave the main process. The renderer makes zero network calls directly. These aren't marketing claims — they're verifiable properties of the code, which is open source on GitHub under the Apache 2.0 license.
Polyphon is free. Not "free tier with limits" — free. You pay your API providers directly for whatever you use. That's the whole model.
Open source
The full source is on GitHub under Apache 2.0. Issues and discussions are open. GitHub Sponsors is enabled for anyone who wants to support development financially. Pull requests aren't open yet, but forks are welcome and the license is permissive.
If you want to support the project today, the best things you can do are download Polyphon, share it with someone who'd find it useful, or leave a star on GitHub. If you want to support it financially, GitHub Sponsors is open.
Polyphon
/ˈpɒl.i.fōn/ · POL-ee-fone
The name comes from polyphony — the musical practice of weaving multiple independent voices together into a single composition. In music, polyphony is what separates a lone melody from a full ensemble. Polyphon brings that same idea to AI: instead of a single model answering in isolation, multiple voices respond, react to each other, and build on what came before. One chat. Many voices.