{"id":11830,"date":"2026-11-17T11:00:33","date_gmt":"2026-11-17T02:00:33","guid":{"rendered":"https:\/\/localmodelwatch.tsuchitsuchi.com\/?p=11830"},"modified":"2026-11-17T11:09:42","modified_gmt":"2026-11-17T02:09:42","slug":"magnitude-cli-0-2-9-released","status":"publish","type":"post","link":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/magnitude-cli-0-2-9-released\/","title":{"rendered":"Magnitude @magnitudedev\/cli@0.2.9 Released"},"content":{"rendered":"<p><!-- lmw:facts --><\/p>\n<h2>At a Glance<\/h2>\n<div class=\"lmw-table-scroll\" tabindex=\"0\" style=\"overflow-x:auto;-webkit-overflow-scrolling:touch;max-width:100%;\">\n<table style=\"width:max-content;min-width:100%;border-collapse:collapse;\">\n<thead>\n<tr>\n<th>Item<\/th>\n<th>Value<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Repository<\/td>\n<td><a href=\"https:\/\/github.com\/magnitudedev\/magnitude\">magnitudedev\/magnitude<\/a><\/td>\n<\/tr>\n<tr>\n<td>Version<\/td>\n<td><a href=\"https:\/\/github.com\/magnitudedev\/magnitude\/releases\/tag\/@magnitudedev\/cli@0.2.9\">@magnitudedev\/cli@0.2.9<\/a><\/td>\n<\/tr>\n<tr>\n<td>Published<\/td>\n<td>2026-10-15<\/td>\n<\/tr>\n<tr>\n<td>License<\/td>\n<td>Apache-2.0<\/td>\n<\/tr>\n<tr>\n<td>Source type<\/td>\n<td>Primary source (the publisher itself)<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<p><em>Values determined by this site&#8217;s code when the information was collected. Dates are JST.<\/em><\/p>\n<p><!-- \/lmw:facts --><\/p>\n<h2>Overview<\/h2>\n<p>The local inference engine &#8220;magnitudedev\/magnitude&#8221; (written in Rust, Apache-2.0 license) has released <code>@magnitudedev\/cli@0.2.9<\/code>. Magnitude is an inference engine designed to auto-tune kernels to match local hardware and run each model at maximum speed, officially claiming speeds twice as fast as Ollama, and operates on Apple Silicon, NVIDIA, AMD, and CPU-only environments.<\/p>\n<p>The most impactful addition in version 0.2.9 is the introduction of server mode via <code>magnitude server setup<\/code>, accompanied by full browser-based UI access. This allows magnitude to run as a persistent background service on headless machines or machines with separate accounts, enabling model downloading, loading, and connection management entirely through a browser. It is worth checking out for those who want to operate without relying on a desktop app.<\/p>\n<h2>Key Changes<\/h2>\n<p><strong>Server Mode and Remote Browser Operation<\/strong> With <code>magnitude server setup<\/code>, you can build a background service that starts automatically at boot, continues running after logout, restarts upon crashes, and performs auto-updates. On Linux, it runs as a dedicated <code>magnitude<\/code> system account as a systemd service, placing data in <code>\/var\/lib\/magnitude<\/code>. On macOS, it is a resident service running prior to login, and on Windows, setting it up as a service via NSSM is guided for remote servers. Once the service is started, accessing http:\/\/<computer>:10100 allows you to use the same Discover, Catalog, My Models, Connections, Usage, Status, and Settings screens from a desktop app directly in your browser. Enabling network access lets you sign in from other devices using a network access key, after which your sign-in state is maintained across reboots and updates. Removal is done via <code>magnitude server remove<\/code>, which preserves models and settings. This is particularly targeted at server operators and users wanting to run models on headless machines.<\/p>\n<p><strong>Connect to Claude Code Without Login<\/strong> It is now possible to connect Claude Code to magnitude without signing in. Requests from Claude Code (including those from built-in sub-agents) are processed using the loaded local model. If you sign in to Claude later, reconnecting will switch to your own Claude login. This is relevant for users utilizing Claude Code as a harness for local models.<\/p>\n<p><strong>Inference Performance Improvements<\/strong> One-off optimization processes have been improved to allocate more time to kernels with heavy actual workloads, and it is reported that generation speeds have increased when running Gemma 4 on Apple Silicon. Furthermore, token packing on Apple Silicon now supports 8-bit weight formats (including DFlash draft models), and packing accuracy for existing formats has been improved, accelerating prompt processing as well. Users running models on Apple Silicon can benefit from this without needing to re-download models.<\/p>\n<p><strong>Model Memory Management and Load Reliability Improvements<\/strong> On Macs, GPU memory for loaded models is now retained even during idle times, avoiding delays in reallocating memory for subsequent requests. Bugs where loading large models caused temporary memory pressure that incorrectly rejected the load have also been fixed, with memory protection now acting only against sustained thrashing or swapping. Additionally, a bug on Linux where available NVIDIA GPU capacity was underestimated at startup (including space occupied by other programs) has been fixed, ensuring model recommendations and compatibility checks are performed based on actual usable capacity.<\/p>\n<p><strong>Installer and Update Reliability Fixes<\/strong> Issues such as package installation failing on glibc 2.34-based Linux distributions like RHEL 9, KDE desktops being unnecessarily installed on headless Ubuntu\/Debian machines, app screens failing to return after a restart on Windows, and leftover installations from other accounts on Mac interfering with app launching have been fixed. Updates that finish installing but fail to launch are now treated as failures on the desktop app side and will no longer retry automatically.<\/p>\n<h2>Supported Models and Hardware<\/h2>\n<p>The following models have been added to the catalog:<\/p>\n<ul>\n<li><strong>Liquid LFM2.5 24B-A2B<\/strong> (Q4, Q5, Q6, Q8): An inference model supporting local agents, coding, and tool use.<\/li>\n<li><strong>MiniCPM-V 4.7 1B \/ 35B-A3B<\/strong> (Q4, Q5, Q6, Q8): Models supporting image understanding and DSpark speculative decoding, including 4x vision projectors and Q8 drafters respectively. Image processing reuses identical crops and checks memory capacity before securing resized images, but does not support 16x visual compression.<\/li>\n<li><strong>Qwen3.8 Flash Next<\/strong> (Q2, Q4, Q6, Q8): A model supporting generation speedup via shared MTP drafting heads. Large n-gram tables are streamed from storage rather than fully expanding into memory.<\/li>\n<li><strong>Qwen3.6 35B A3B Q2 Edition<\/strong>: A smaller variant supporting DFlash speculative decoding.<\/li>\n<\/ul>\n<p>Existing users can use these newly added models by downloading them. Re-downloading is only required for newly added models, and re-conversion of existing models is unnecessary.<\/p>\n<h2>How to Get It<\/h2>\n<p>Install or update via the terminal using the following commands:<\/p>\n<p>macOS and Linux:<\/p>\n<pre><code>curl -fsSL https:\/\/magnitude.dev\/install.sh | sh\n<\/code><\/pre>\n<p>Windows (PowerShell):<\/p>\n<pre><code>irm https:\/\/magnitude.dev\/install.ps1 | iex\n<\/code><\/pre>\n<p>On Linux, the installer will ask once at startup whether to run Magnitude as a server (default is &#8220;No&#8221;, and it is automatically skipped if there is no terminal or no response within 60 seconds). Passwordless sudo environments will not prompt for a password.<\/p>\n<p>If you are already using the desktop app or <code>serve<\/code>, updates can be applied by clicking &#8220;Restart&#8221; in the desktop app or using the following command:<\/p>\n<pre><code>magnitude update install\n<\/code><\/pre>\n<p>You can check for updates with <code>magnitude update check<\/code>. If you want to run it persistently as a server, run the following after installation:<\/p>\n<pre><code>magnitude server setup\n<\/code><\/pre>\n<p>To remove the server, use <code>magnitude server remove<\/code>, which keeps your models and settings intact.<\/p>\n<p><!-- lmw:related --><\/p>\n<h2>Related Articles<\/h2>\n<ul>\n<li><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/10\/06\/magnitudedev-magnitude-026-released\/\">magnitudedev\/magnitude v0.2.6 Released: Major Mac Speedups<\/a><\/li>\n<li><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/magnitude-cli-0-2-14-released\/\">Magnitude CLI 0.2.14 Released: Faster CPU Decoding and More<\/a><\/li>\n<li><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/magnitude-cli-0-2-11-released\/\">Magnitude CLI 0.2.11 Released with Custom GGUF Support<\/a><\/li>\n<li><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/magnitude-cli-0-2-10-released-2\/\">Magnitude CLI 0.2.10 Released with Memory Fixes and New Quantizations<\/a><\/li>\n<\/ul>\n<p><!-- \/lmw:related --><\/p>\n<p><!-- lmw:next-steps --><\/p>\n<h2>What to Read Next<\/h2>\n<ul>\n<li><strong>Follow this tool<\/strong> \u2192 <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/engine-magnitude-en\/\">Magnitude overview and release history (16 releases tracked)<\/a><\/li>\n<li><strong>Other inference engines and runtimes<\/strong> \u2192 <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/engine-llama-cpp-en\/\">llama.cpp<\/a> \/ <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/engine-ollama-en\/\">Ollama<\/a> \/ <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/engine-vllm-en\/\">vLLM<\/a><\/li>\n<\/ul>\n<p><!-- \/lmw:next-steps --><\/p>\n<h2>Sources<\/h2>\n<ul>\n<li><a href=\"https:\/\/github.com\/magnitudedev\/magnitude\/releases\/tag\/%40magnitudedev%2Fcli%400.2.9\">magnitudedev\/magnitude @magnitudedev\/cli@0.2.9 Release Notes<\/a><\/li>\n<\/ul>\n<h2>Sources<\/h2>\n<ul>\n<li><a href=\"https:\/\/github.com\/magnitudedev\/magnitude\/releases\/tag\/%40magnitudedev\/cli%400.2.9\">https:\/\/github.com\/magnitudedev\/magnitude\/releases\/tag\/%40magnitudedev\/cli%400.2.9<\/a><\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>Magnitude 0.2.9 is released, featuring server mode setup, web UI support, Claude Code connection, and performance improvements.<\/p>\n","protected":false},"author":1,"featured_media":11829,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[105],"tags":[3408,2791,3057,2793,1547],"class_list":["post-11830","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-engines-and-tools","tag-local-inference-en","tag-magnitude-en","tag-magnitudedev-en","tag-rust-en","tag-verified"],"lang":"en","translations":{"en":11830,"ja":11828},"_links":{"self":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts\/11830","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/comments?post=11830"}],"version-history":[{"count":2,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts\/11830\/revisions"}],"predecessor-version":[{"id":17791,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts\/11830\/revisions\/17791"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/media\/11829"}],"wp:attachment":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/media?parent=11830"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/categories?post=11830"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/tags?post=11830"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}