Pheron Agent runs entirely on Apple Silicon — planning, tool use, and self-correction across 72 integrated tools, with nothing leaving your machine. Built in Swift, tested against real bugs, not a demo.
Hey — I'm Turgay, I built Pheron Agent alone, mostly at night, over the last several months.
The short version: it's a macOS app that runs a full agent loop — plan, call tools, read the result, correct itself — entirely on your own Mac. No API key, no account, no data leaving the machine. It talks to 72 tools (files, shell, browser, calendar, Jira, Slack, Notion, and a lot more) through a real Swift actor architecture, not a Python wrapper around someone else's API.
Every one of those 72 tools ships behind a real test suite — 369 tests right now — and I don't merge a fix without running it against a live account, not a mock. Every MCP integration (Jira, Slack, Notion, GitHub, and more) was connected to a real account and driven with a real prompt before I called it done. The full testing methodology (300+ real scenarios, not synthetic benchmarks) is public: https://github.com/trgysvc/Agent...
I didn't build this to prove a point about local AI. I built it because I stopped mid-paste one day — about to hand a year of product strategy to a chat window — and realized I never had a real answer for "where does this actually go." I wrote the longer version of that here: https://trgysvc.github.io/blog/i...
A couple of weeks ago I found a bug where it told me a Jira ticket move had failed, and the ticket had actually already moved — the agent ran out of "turns" one step after it succeeded, and reported the timeout as a task failure. Wrote that one up too: https://trgysvc.github.io/blog/t...
Pricing is $69 one-time (normal price is $99, this is still the launch discount), one Mac license, 14-day no-questions refund. Needs Apple Silicon and macOS 26+.
Happy to answer anything about the architecture, the sandboxing, or why I made specific calls the way I did — I'll be around in the comments all day.