Skip to content

EZLM 0.1.12 (beta)

Latest

Choose a tag to compare

@Picom-code Picom-code released this 06 Nov 22:04
· 3 commits to main since this release

Connect now sets up OpenCode, Hermes and DeepSeek Harness for you, checks that each one really works with your model, and opens them in one click.
Updating keeps your models, keys, settings and chats. Press Update in Settings.

This release is for Windows (regular and ARM PCs). The Mac app stays on 0.1.11 for now; a Mac build of these changes follows.

What's new in 0.1.12

  • Connect, rebuilt. Pick an app, press Connect, and EZLM:
    • installs the app if you don't have it, at a tested version, and checks the download before using it;
    • adds your model to the app as "EZLM", without touching your other settings;
    • asks the app to read a file with one of its tools and answer, so "Connected" means it really works;
    • shows each step as it goes, and a plain reason if something fails, with Show details and Report a problem.
  • DeepSeek Harness. A third app on Connect. Open it in your browser, or with Open in terminal in a terminal app (dsh-TUI), which EZLM installs the first time.
  • Open buttons on Overview. Open OpenCode, Hermes or DeepSeek Harness straight from Overview. EZLM starts your model first if needed, and tells you if an app closes right after opening, and why.
  • OpenCode:
    • EZLM uses your own OpenCode when it works with your chat history, and otherwise installs its own OpenCode 2.2.5 (an older OpenCode can't open the history a newer one wrote).
    • A first answer that took minutes on small thinking models now comes in seconds (chat titles no longer make the model think).
  • Hermes:
    • Works with Hermes installed from its script or from the Microsoft Store, and never installs Hermes's background service or a login item.
    • If Hermes needs a larger context than your model has, the tile says so and Connect again makes the room.
    • The first check on a new Hermes can take a few minutes while Hermes sets up its tools; Connect now says so.
  • Memory. EZLM now counts the memory your model's conversations keep between turns, and gives the rest to llama-server's prompt cache instead of a fixed 8 GB. A full cache can no longer push your computer into swap. On a few computers this picks a different model than before (for example, a 12 GB graphics card now gets a model that fits on the card instead of one that pages from disk).
  • Clearer progress. Installer warnings that don't mean anything is wrong stay in the log instead of showing as progress, and the first-answer estimate is closer for waits around a minute.
  • Windows on Arm. OpenCode, Node.js and DeepSeek Harness install as native Arm versions.
  • Overview lists the apps you've connected, and Settings > About credits the apps' makers.

Install

  • Windows 10 or 11 (regular and ARM PCs): download EZLM-Setup.exe and open it. If Windows says "Windows protected your PC", click More info, then Run anyway.
  • Mac with Apple Silicon (M1 or later): this release has no Mac build. Use EZLM 0.1.11 from https://github.com/picom-code/ezlm/releases/tag/v0.1.11-beta for now.

Download page: https://picom-code.github.io/ezlm/

If something goes wrong

Press Report a problem in EZLM (bottom left, on the error itself, or on an app's tile). It opens a filled-in report on EZLM's public GitHub page; press Submit new issue there. Your keys and files are never included.

SHA-256

  • EZLM-Setup.exe ED9B10A2B80120A9049D2CA6C7686ECE26F6370258DEBE3B2432C5490E230FD0