Workshop / NamerKit

NamerKit

A naming studio for macOS: a SwiftUI app sends your brief to local or remote language models, filters what they propose and keeps the reasons; its newest mode skips ranking and shows unranked waves of names to sort by hand.

State: 5 Oct 2026, private and in development: a first “waves” mode was added after two automatic ways of predicting the author’s taste did no better than chance; no benchmark against Namelix has yet passed. m

commits per week, 12 weeks · 0 in the last 7 days · as of 14 Oct

Runs on: macOS 14 or later; local models need an Apple Silicon Mac with 21 GB or more
Built with: Swift 6 and SwiftUI, a Python 3.11 engine, MLX runtimes and a LiteLLM gateway
Data: Local files; a remote run sends the brief to its provider
Code: Private repository

Known limits

No comparison with Namelix has passed: 46 % wins on 10 briefs in July, below the 50 % target; a 74 % result was set aside in September. m
Two automatic ways to pick names to the author’s taste failed their pre-registered tests; the author still sorts by hand. m
The default run is remote; the local models in use weigh about 21 to 177 GB. m
Not yet distributable: the app needs the Python engine and downloaded models. m

Sources and questions

Inquiry: Michael Levin, spaces of possibility, the geometry of models, and the limits of a Platonic reading

Register · 7 entries

05 Oct measure
Two automatic ways of predicting the author’s taste, a base model’s probabilities and a language-model judge, do no better than chance on pre-registered tests; the studio moves to unranked waves sorted by hand. m
05 Oct measure
Prompting for distant territories cuts obvious junk names from 16% to 3% but the author keeps 0.2 names per 10 instead of 1.8, so the approach is dropped. m
26 Sept correction
A 74% win rate recorded in July is set aside: the name format gave the source away, the Namelix sample was reused and the test briefs were the tuning briefs. c
07 Jul measure
A blind scoring by one person gives 46% wins against Namelix on 10 briefs, up from 34% in May; the 50% target is missed. m
08 May decision
The on-device model is dropped for a naming studio: local language models propose names, and filters and scores sort them. m
04 May correction
Scored again by an agent that had not seen the data, the first blind benchmark against Namelix falls from a ratio of 0.84 to 0.74 and fails; a report that printed a pass is fixed the same day. m
02 May milestone
NamerKit starts as a plan to match Namelix with a small naming model trained from scratch and run on the device. m