Should you try LingChat?
Try it when the character presentation is the reason you want an AI companion. The README describes independent save memories, custom characters, changing expressions and backgrounds, multi-character scripts and desktop-pet mode. Those are project-described capabilities, not behaviors verified by our sandbox.
Our run on September 27, 2026, at commit 848fd34 recorded successful install and build commands, but skipped tests because no test script was detected. That supports proceeding to a manual evaluation; it does not establish everyday reliability or security.
What our sandbox established
The stored run used the lab-node:22 image with 3 CPUs and 8 GB of memory. Installation took 17 seconds and the build command took 39 seconds. Both records contain success status, exit code 0 and no timeout. The harness keeps elapsed time and output tails alongside those command outcomes, and records a skipped test step separately from a passing or failing suite.
The build output ends with generated frontend assets and a warning about large minified chunks. Its internal "built in 11.99s" footer covers a narrower operation than the harness's 39-second build step; it does not replace that stored duration. Likewise, the install tail says "Packages: +253", while the stored dependency measurement is 254 packages. These are different recorded figures, so we do not substitute the log's added-package count for the dependency measurement.
No automated test result was produced. We did not exercise model conversations, persistent memory, script choices, audio, screenshots, tool execution, desktop resizing or mobile use. The audit field is null, so there is no recorded advisory result and scan coverage is unknown.
The MrKeyoor testing-methodology page explains command records and their limits. It is background for interpreting this stored run, not evidence that September's run used today's procedure or that we executed LingChat again.
Configuration and permissions need a manual check
The README asks users to supply model-provider credentials. Its primary setup instructions are Chinese. The v0.5.0 release says the application supports simplified Chinese, traditional Chinese, English and Japanese; interface translation does not establish equivalent documentation in each language.
LingChat v0.5.0 describes configurable character tools that can operate the computer and search the web. The release notes also describe skills, plugins, script generation and built-in TTS. Our build result verifies none of those interactions.
The README documents screen observation triggered through conversation and proactive observation after configuration. Before enabling those features, check which model receives images and which tools the character can call. A useful first evaluation is to configure a disposable character, exchange messages, restart to inspect memory, try a script choice and test voice. Then evaluate screen observation and individual tools with content you are comfortable sending to the configured provider.
The reported UI problems have a later resolution
Issue 707 is closed, and the maintainer reported the raised UI problems fixed on October 14, 2026. The original report concerned v0.5.0 on Windows 11 and described missing controls, transparent areas and misplaced elements after resizing or zooming. Its later discussion records fixes, with the maintainer qualifying that extreme zoom is not guaranteed to display normally.
That resolution is source evidence from after our September run. We did not verify the fixes, and the old report should not be presented as proof that those defects remain current. Check your usual window size and zoom during a manual trial.
Choose an alternative by the interaction you need
SillyTavern is a locally installed interface for text-generation models, image-generation engines and TTS voice models. Its README also lists Visual Novel Mode, lorebooks and extensions. Consider it as a replacement when configuring model interactions and prompts matters more to you than LingChat's particular character and script workflow. Visual-novel presentation alone does not distinguish the two.
AIRI documents Live2D and VRM support, browser audio input and multiple voice-synthesis providers. Its README also calls the project early in development and identifies several capabilities as work in progress. Consider it as another companion platform when those avatar and audio capabilities match your plans. The extracts do not establish that it is more reliable or easier to install than LingChat.
These are alternative applications, not required LingChat components. A VITS service, by contrast, can complement LingChat by providing speech synthesis.
Code permission does not settle asset permission
The recorded checkout identifies AGPL-3.0 as the code license. The README separately attributes some interface and audio materials to games and prohibits commercial use; it also restricts commercial use of the default character artwork. Review those permissions separately before distributing a commercial derivative.
How to read the scores
The ratings are editorial judgments, not benchmark results. On this five-point rubric, 1 means substantial documented obstacles, 2 means important validation gaps, 3 means a usable documented path with configuration or language friction, 4 means documented participation and response, and 5 requires broader corroboration than this evidence supplies. Each dimension uses only the relevant evidence: setup instructions and command outcomes for setup, available guidance for documentation, visible participation for community, and validation limits for maturity. Community responsiveness does not measure adoption or guarantee support.
