Devotd Github
NewModel Context Protocol

itsmostafa/system-one-connector 0.4.14 — Fixes two evaluation bugs

itsmostafa/system-one-connector released 0.4.14, incorporating two eval bug fixes: one that switches the taxcat evaluation to a neutral state, and one that displays recall separately per group.

Repositoryitsmostafa/system-one-connector ↗

itsmostafa/system-one-connector released 0.4.14, incorporating two eval bug fixes: one that switches the taxcat evaluation to a neutral state, and one that displays recall separately per group.

GitHub description: System One MCP connector to evaluate anything fast and cheap. Give your AI agent direct access to models like: Jev, D1, CLM and Laya

stars
★ 352
Forks
41
Language
Go
License
MIT
Latest release
v0.4.12
Read original source ↗View on GitHub ↗

A 0.4.14 release was posted for itsmostafa/system-one-connector. This repository is an MCP (Model Context Protocol) connector that directly connects System One models to AI agents, and the repository description states that it lets agents access models such as Jev, D1, CLM, and Laya directly. This commit (33d1fc68, 2026-11-03) is a release-please automation merge, and rather than the code itself it only bumps the version manifest from 0.4.13 to 0.4.14 and adds fix entries to the CHANGELOG.

What is listed for this release is two bug fixes in the eval area. The first switches the evaluation named taxcat to a neutral state and makes it score only against its own dataset, and the second makes recall (the proportion of the correct answers that were found) be displayed separately for each taxcat group, whose evaluation limits (limits) differ. Both are fixes that change only how metrics are computed and reported, and they do not touch the model access functionality itself. According to the CHANGELOG, the release date is 2026-10-25, and the two fixes are linked to the commits 12212ee and 2b0f555.

The judgment stops here. The presented change list consists of two files, the version manifest and the CHANGELOG, and the commit statistics are on the order of 9 lines added and 1 line deleted. Exactly how the evaluation logic was fixed can only be confirmed by reading the two commits that the CHANGELOG points to directly. The performance figures or benchmark results for the models listed in the repository description, and the license information, are not present in the provided facts. To decide whether to adopt it, it appears better to check the eval script and the license directly and then make the decision.