Aximo — a local Rust STT API for CPU-only inference

Dmitry — Wed, 22 Apr 2026 22:20:47 +0000

I built a local speech-to-text API in Rust that runs on CPU

I recently built Aximo, a self-hosted speech-to-text microservice designed to run locally on CPU, without depending on cloud APIs or external SaaS.

The idea was straightforward: I wanted an STT service that could be deployed like any other backend, stay fully local, and still be clean enough architecturally to evolve beyond a quick experiment.

Aximo is written in Rust, uses Parakeet v3 for local inference, exposes an HTTP API for transcription, and includes a WebSocket layer for realtime use cases. I also added Docker, OpenAPI, and a multi-crate workspace layout to keep the codebase modular from the start.

One detail I particularly liked: I extended Swagger UI so I can record audio directly from the microphone and send it to the API for testing. It’s a small feature, but it makes the developer experience much nicer when iterating on the service.

At this point, I’d call it a solid MVP rather than a production-ready system, but it already works well for local experimentation and as a foundation for a self-hosted STT stack.

One notable addition: I extended Swagger to support sending recordings directly from the microphone.

Repo: github.com/aximo

Secure AI Agent Architecture

Dmitry — Sun, 29 Mar 2026 10:35:30 +0000

I’ve Started Writing an Open Book on Secure AI Agent Architecture

I’ve started writing an open book on the architecture of secure AI agents.

The goal is to build a practical engineering reference — not a collection of flashy demos, but a structured guide to production-grade agent systems: control planes, policy boundaries, tool execution, memory, observability, evaluations, approvals, and governance.

The first chapters are already live:

English: https://agent-axiom.github.io/agent-arch/en/
Chinese: https://agent-axiom.github.io/agent-arch/zh/

Repository: https://github.com/agent-axiom/agent-arch

There is a lot of excitement around agents, but far less shared engineering guidance on how to build them safely and operate them reliably in production. This project is my attempt to help close that gap.

I’d genuinely appreciate thoughtful feedback from the community:

what feels solid
what is missing
what seems debatable
what should be improved
what operational or security practices deserve more attention

If this topic is close to your work, I’d be glad to hear your critique, ideas, counterexamples, and contributions.

Forem: Dmitry

Aximo — a local Rust STT API for CPU-only inference

Secure AI Agent Architecture

I’ve Started Writing an Open Book on Secure AI Agent Architecture