Install

LensBench comes in three forms: the desktop app (the main one, installed on your own computer), the web version (use it straight from a browser), and MCP (lets your own AI client operate lenses). The desktop app and MCP come from the same installer; the web version needs no installation.

Desktop app

System requirements: Windows 10 / 11 (64-bit). macOS and Linux are not available for now.

1. Download: get the latest portable build (about 320 MB, with a SHA-256 checksum) from the pinned download post on the forum, LensBench desktop download.

   The download link lives in that one post only and is replaced there with every release, so it is always the latest. This is currently a test version, shipped as a portable build (unzip and run); there is no installer program yet.

2. Unzip and run: unzip into a folder (for example D:\LensBench) and double-click lensbench-desktop.exe inside. There is no installer, so nothing is added to the Start menu and .lens files are not associated automatically; those come with the installer of the official release. The program is not code-signed, so Windows may show "Windows protected your PC": click "More info" → "Run anyway". Some antivirus programs may raise a false alarm; add the unzipped folder to the trusted zone. Sign-in required: the first time you open it, you must sign in with an optic.chat account before you can compute (a new device gets a 7-day full-featured trial).

3. Open: you start on the home page, where you can create from a template (single lens / Cooke triplet / blank), open a .lens, import a Zemax .zmx, open a reference design, or open a recent file. Next, read Quickstart and the GUI guide.

The install directory also contains the command-line program lb.exe (the same command table and the same engine as the desktop app, suited to scripts and batch jobs; see the command reference) and the MCP server lensbench-mcp.exe (see "MCP" below).

Updating

After startup, the desktop app checks for a new version automatically (at most once a day). When there is one, a notice appears above the analysis view:

1. Click "更新…" (Update…) to see the version number and release notes, then click "下载" (Download) to download the installer to your machine (nothing is downloaded until you click).
2. After the download it is verified automatically; once verification passes, click "安装" (Install). Installing requires quitting the program: if there are unsaved designs, they are listed and you are asked to choose "放弃修改并安装" (Discard changes and install) or "取消" (Cancel) (cancel first if you want to keep your changes, save, then install). Choosing Cancel skips the update and the program stays open as usual; after installing, it reopens automatically. During installation, AI clients connected to LensBench (MCP) are also disconnected; just restart the AI client.
3. If you don't want to update right now, click "稍后" (Later); if you don't want this version, click "跳过这个版本" (Skip this version) (later releases are still announced as usual). A version marked "required update" cannot be skipped.

You can always check manually with Help "?" › 检查更新… (Check for updates…) (skipped versions are shown as usual).

• The update check only sends a GET request to optic.chat, carrying the software version number; it does not include your designs, usage data or machine information. For when else the software goes online and what it sends, see Privacy and network access.
• In Preferences (the ⚙ in the top bar) you can turn off the automatic check (after that it no longer checks at startup; check manually when you want to), or switch to the "beta" update channel (which also receives stable releases). Switching from beta back to stable does not downgrade: the beta you have installed keeps working, and you are prompted to update only when a stable release newer than it comes out.
• Version information carries a digital signature and the installer carries a checksum: if the signature is wrong, or the installer differs from what was released, it is never installed.
• Command line: lb update shows whether there is a new version and its release notes; lb update --download downloads the installer and verifies it (and tells you where it is; close the program and run it to install).

You can also download a new installer from the website at any time and install over the existing one.

Web version

The web version is paused while it moves onto the same computation engine as the desktop app, and it will reopen once that is done (web.optic.chat currently shows a "coming soon" page). Please use the desktop app in the meantime. Existing optic.chat accounts still work: sign in to the desktop app with the same email and password.

MCP: let your own AI client operate it

The MCP server is installed together with the desktop app; no separate installation is needed.

1. Open the desktop app and click "AI" at the right end of the ribbon (Ctrl+Shift+A) to open the "AI 协作" (AI collaboration) panel.

2. Under "连接你的 AI 客户端" (Connect your AI client), choose a client (Claude Code, Claude Desktop, Cursor, Cherry Studio, Trae) and click "写入配置文件" (Write config file) (it backs up the original file first and leaves other servers alone), or "复制配置" (Copy config) and paste it yourself (Cherry Studio supports only this: use "Import from JSON" in its settings). The project directory is the folder containing the current lens: the AI can read and write only files in that folder.

3. Restart the client as the panel instructs, then just say, for example:

   Open my.lens and change the thickness of the last surface to a paraxial focus solve, and see how much the total length changed.

   Before the AI changes a design, it shows you the diff and the before/after metrics, and writes only after you confirm; in the desktop app's "AI 协作" panel you can see that it is online and which tool it just called.

For differences between clients (timeouts, progress, tool-count limits) and the command-line way of writing the configuration (lb mcp-config), see Native AI.