NanoTaste

Local research harness · v0.1.0-rc0

Install it. Answer a few prompts. Let it keep pulling the sessions that train the file.

NanoTaste is a local CLI and studio for a hierarchy of taste files. On first run it discovers your coding agents, asks how often to harvest sessions, and can seed the files from a personal site, pictures, or notes. It has not demonstrated human preference alignment.

Watch the flow Install in four commands Read the docs

What you get on first install

  1. Setup asks the useful questions once. Which agents to integrate, which domains to keep, and how often to pull sessions.
  2. A main taste file with category children. Writing, code, aesthetic, product, personal, brand, communication, and research.
  3. A seed box for your own material. A personal website, files, images, or pasted notes become inspectable overlays.
  4. A scheduled harvest. Daily, weekly, or monthly pulls refresh learned overlays without silently rewriting the seeded files.

Local studio

After install, nanotaste serve opens a localhost app. You can see the hierarchy, tags, what the files currently encode, add seeds, and run a harvest. It is not this marketing site.

Marketing site

This site is the public explanation: the install loop, the video, the docs, and the claim boundary. It does not read your sessions or write taste files.

Get started

python3 -m venv .venv
. .venv/bin/activate
python -m pip install .
nanotaste setup

Non-interactive install with a first seed and weekly harvest:

nanotaste setup --yes --every weekly --seed-url https://example.com/about
nanotaste serve

Honest limit

The harvest extracts lexical signals. Scheduled overlays live in taste/learned/. Seeded category files stay reviewable. NanoTaste does not train a preference model, and has not demonstrated human preference alignment.