Feyer

Paul Feyerabend, Against Method

And the initial playful activity is an essential prerequisite of the final act of understanding... Creation of a thing, and creation plus full understanding of a correct idea of the thing, are very often parts of one and the same indivisible process and cannot be separated without bringing the process to a stop. The process itself is not guided by a well-defined programme, and cannot be guided by such a programme, for it contains the conditions for the realization of all possible programmes. It is guided rather by a vague urge, by a [Kierkegaardian] passion. The passion gives rise to specific behaviour which in turn creates the circumstances and the ideas necessary for analysing and explaining the process, for making it rational.

Part play, part product; part research output, part tool I use to write and research; Feyer is an AI-native workspace application designed and built for both humans and agents, providing people like you and me all the UX/UI niceties while provisioning agents with the tools and data structures they need to be effective.

As LLMs increasingly operate as agents, working across and within different digital environments, I was curious to explore how you might go about designing and building double-facing interfaces that can serve both humans and agents. Feyer is then both the ongoing product of this exploration as well as the tool I use to gather research, run benchmarks, and capture insights from everyday use, to guide where I might want to go next.

Features and Explorations

Lean in-app AI, robust MCP and agent tooling

Feyer packages lean, focused, in-app AI features while providing robust MCP and agent tooling allowing for agents from Claude, Codex, and other model providers, to easily work alongside you within Feyer.

Across devices · Demo 1

Coming soon

Apps for Mac, iPhone, iPad, and MCP

Some surfaces better accommodate specific activities or contexts. I like to write on my desktop, read and highlight papers on my iPad, and be able to capture random thoughts in my day to day. Feyer can be used across all three surfaces as you see fit.

Want to read a paper within Claude or Codex, asking questions as you go? You can! Want to use ChatGPT’s voice agents to talk through your research or refine a concept aloud? That too! All the surfaces are there to let you explore an idea from every angle.

Research library · Demo 1

Coming soon

Large, embeddable, indexed library

Bring your work in from Zotero with your existing highlights preserved, or pull a paper in by DOI, arXiv, OpenAlex, and Crossref; clipped pages arrive as readable PDFs rather than scraped text. Everything is indexed on the way in and retrieval stays lexical throughout, so what you search for is what you actually get back.

Because the library sits behind the same MCP and agent tooling as the rest of Feyer, it travels with you. An agent can open the original PDF at the page it cited, its passages on one layer and your own annotations on another, meaning the sources you gathered stay useful wherever you happen to be working.

Authorship · Demo 1

Coming soon

Lean in the app, deep at the agent boundary

In the app itself there are three narrow commands, and the tightest of them is capped at phrase length on purpose. At that grain you can hold every option in mind at once, so choosing is still authorship; past it, choosing becomes delegation. There is no chat panel.

The depth sits at the agent boundary instead. Tools carry provenance on every write, refuse to edit what you wrote, and can author reports but never your prose. A delegated read returns the source’s own bytes, and a quote that cannot be located comes back as an orphan rather than a guess.