
Introducing feros
Why we built feros and how to run an open, airtight voice AI stack in your own infrastructure.
We built feros Voice Agent OS to solve a practical problem: shipping production voice AI without giving up control, spending weeks on pipeline plumbing, or being locked into per-minute platform economics.
feros is fully open source on GitHub: github.com/ferosai/feros.
feros is designed to be open, airtight, and enterprise-grade. The platform combines:
- A Rust voice runtime for sub-second call handling
- An AI-driven builder for agent setup
- A Python control plane for configuration and operations
- A self-hostable monorepo that you can run in your own infrastructure
The architecture is intentionally split into two independent layers:
- Control plane (Python): config, management, integrations, evaluation
- Voice runtime (Rust): live call orchestration across VAD, STT, LLM, and TTS
That separation gives teams flexibility: swap model providers, telephony vendors, and deployment targets without rewriting the whole system.
If you want to try it quickly:
git clone https://github.com/ferosai/feros.git
cd feros
cp .env.example .env
docker compose up -dThen open http://localhost:3000 and start building.
This blog will track what we ship next: runtime improvements, integrations, evaluation workflows, and production hardening notes from real deployments.
If feros is useful for your team, please consider giving the project a star on GitHub to help more builders discover it: github.com/ferosai/feros.