Voice Agents
Platform for prebuilt "Digital Employee" agents that answer phone calls and chat, configured rather than prompt-engineered, and running on any Kubernetes cluster.
Project Details
Client
Agentic Human
Year
2026
Role
Co-Founder and sole engineer
Status
Live beta
Ownership
Co-founder and sole engineer at Agentic Human
Technologies
The Challenge
Businesses want an agent that picks up on the first ring, uses their tools, and bills predictably, without sending audio to third-party speech APIs or writing prompts from scratch.
The Solution
A Rust gateway takes calls from Twilio Media Streams and Exotel, plus text over WebSocket and SSE, and runs voice activity detection, speech-to-text, and text-to-speech on-device behind cargo feature gates. A Rust agent runner orchestrates sessions with any OpenAI-compatible LLM the customer brings. User tools and platform connectors run in a deno_core isolate pool. Agents ship as versioned templates that must pass functional and adversarial evals. Billing is a prepaid wallet over append-only Postgres ledgers.
Highlights
- On-device speech pipeline with no external speech APIs
- Sandboxed tool and connector execution in V8 isolates
- Portable Kubernetes design running unmodified on DOKS, GKE Autopilot, and EKS
Outcomes
- Public site at agentichuman.io