Introducing Fervon: a one-person software studio run by AI agent fleets

I've shipped a lot of small products this year — scattered across repos, names and landing pages with nothing tying them together. So I gave them a home: Fervon.

Fervon is a software studio. The unusual part is how it works: it's one builder running fleets of AI agents. I design the product and the architecture, then orchestrate agents to implement, review and ship — often several projects in parallel. Think of it as a small software factory with a single human at the helm.

What's in the forge

Trace — a local-first personal memory app. It captures lightweight signals from your browsing and activity and lets you search everything and scroll a timeline of your digital life: 100% on your machine, no screen recording, no cloud. (The "Rewind that Meta killed", minus the creepy part.)

The open-source tools:

• inferbench — download, launch and benchmark local LLM engines from one desktop app. Real tok/s on your own GPU, no simulated numbers.
• ClaudeScope — a local dashboard plus full-text search over your Claude Code sessions. Zero deps, zero network.
• Lookspan — lightweight, local-first observability for AI agents (spans and traces).
• Launchpad — a local launcher that discovers and runs all your projects on unique ports, no collisions.
• Pregón — a cross-poster that adapts one update to every social channel. (This very article was published through it.)

Why "Fervon"?

From the Latin fervere — to burn, to boil, fervor. The whole brand is built around the forge: things come out fast, hot, and ready to use. The tagline is "Forged red-hot."

Where it's going

Fervon is a "house of brands": the studio is the label, and each product keeps its own identity (e.g. Trace by Fervon). Free and open tools stay free; a couple of products — starting with Trace — are paid and fully self-serve, no sales calls.

If you like local-first software, AI tooling, or watching someone build in public with an unusual workflow, come along:

👉 https://fervon.dev · code at github.com/fervon

Happy to go deep in the comments on how the agent-fleet workflow actually works.