SolonCode v2026.8.4: Adjustable UI Fonts, 22 Languages, and Smarter Memory Search

#solon #java #ai #tools

A fully Chinese-capable coding agent that goes to work the moment you open a terminal. This release rounds out three things at once: read it clearly, use your own language, and actually find what it remembered.

What is SolonCode

SolonCode is an enterprise-grade terminal coding agent built by Hangzhou Noear Technology — a digital teammate that understands requirements, plans steps, and writes code on its own. It's not picky about models or platforms, and a single curl line gets you running.

Repos: Gitee · GitCode · GitHub

Install

Runs on Java 8+ (8~26), across macOS / Linux / Windows / Harmony PC.

# macOS / Linux / Harmony PC
curl -fsSL https://solon.noear.org/soloncode/setup.sh | bash

# Windows (PowerShell)
irm https://solon.noear.org/soloncode/setup.ps1 | iex

For the first run, start soloncode web 0 (0 picks a random port; you can also pin one, e.g. soloncode web 1212), then head to Settings → Models to add a model and test the connection.

Three reasons to upgrade

1. Adjustable UI fonts — no more squinting through long sessions

Settings → General → Fonts adds three independent controls: UI Font (12 searchable options), Code Font (independent from the UI font), and Font Scale (85%–150%, 5% steps, with a one-click reset).

Changes apply instantly, no restart needed. The scaling decouples type size from layout — when you enlarge text, fixed-size elements like icons, borders, and control heights don't get stretched out of shape. That's something full-page browser zoom could never do cleanly.

2. 22 UI languages — overseas developers land on their native tongue

Settings → General → Display Language offers Follow System + 22 languages, covering Simplified and Traditional Chinese, English, Japanese, Korean, German, French, Spanish, Italian, Russian, and Portuguese — plus less common ones like Arabic, Bengali, Bosnian, and Norwegian. It's not a token English-Japanese-Korean gesture. The default follows your system, so overseas developers land on a native interface right after install.

Two notes: the UI language is independent from the model's reply language — switching to English only changes the interface text. And in this version the language choice isn't persisted yet (a refresh reverts to Follow System); this has been reported to the core team.

3. Memory — from "can store" to "can find again"

Memory took shape across four releases: v2026.7.27 added the web management panel, v2026.7.28 introduced scopes (shared cross-project knowledge goes to "User Global," project-specific conventions go to "Current Workspace," and switching projects keeps them from interfering), and v2026.8.4 refined the prompting and search.

Search targets a real pain point: once you have hundreds of entries, "stored but unfindable" is as good as not stored. Now you can browse by importance and time, and use semantic retrieval to pinpoint what you need.

While we're at it: engineering touch-ups

The web workbench keeps growing into a serious tool: sub-agent management (v2026.7.24, four built-ins — bash / explore / general / plan, none with write access except general; drawing that boundary in config is more reliable than leaning on prompts), context-compaction threshold as a ratio (v2026.7.26, default 75%, no recalculating when you swap models), a simplified /goal command, an enhanced Git review panel (diff/stat, rollback, directory detection, garbled-Chinese fix), config reload (no restart across instances), skins, and the continuously iterating desktop build.