WEILI SHI:<\/strong> Let\u2019s dive into MagenticLite.<\/p>\n\n\n\nYou may have heard of Magentic UI, the agentic application we released last year that set the foundation for this work. With MagenticLite, we reworked the agent harness to run efficiently on small-language models\u2014making it faster, more lightweight, and no longer reliant on frontier-scale models.<\/p>\n\n\n\n
We also refreshed the UX design based on community feedback, making it easier and more natural to work with.<\/p>\n\n\n\n
MagenticLite works across both the browser and your local file system to help you get real work done\u2014whether that\u2019s filling out online forms, making appointments on your behalf, managing files on your desktop, or generating simple code.<\/p>\n\n\n\n
Let\u2019s see it in action.<\/p>\n\n\n\n
I can give MagenticLite access to a folder on my computer. Here, I\u2019m giving it notes from the last Microsoft Build conference. I want the agent to search for what has changed and create an update document to help me prepare for the next conference.<\/p>\n\n\n\n
It has successfully accessed my notes and created to-dos. Next, it opened a web browser and started to gather the information I need.<\/p>\n\n\n\n
MagenticLite has access to its own browser running in a virtual machine. This helps minimize the risk of data leakage while allowing Fara, our browser-use model, to operate quickly.<\/p>\n\n\n\n
Fara performs well at long-running tasks. It looks like Fara has gathered enough information, and the Orchestrator has created a document. Let\u2019s check it out.<\/p>\n\n\n\n
MagenticLite did the job. It included updates on all key sections.<\/p>\n\n\n\n
Next, I\u2019d like to email this document to my colleague. I\u2019ll let MagenticLite do this for me. I can keep working on other things, and MagenticLite will notify me when my attention is needed.<\/p>\n\n\n\n
In this case, it needs my help to log into my email account. I\u2019m taking control of the browser to log in.<\/p>\n\n\n\n
Once unblocked, the agent composes the email and can send it once it finishes.<\/p>\n\n\n\n
Now that you\u2019ve seen it in action, let\u2019s take a closer look at the models powering MagenticLite\u2014small models that punch above their weight. Next, Harkirat will introduce Magentic Orchestrator.<\/p>\n\n\n\n
HARKIRAT BEHL:<\/strong> Thanks, Weili.<\/p>\n\n\n\nIf MagenticLite is the app you interact with, and Fara drives the browser, then Magentic Orchestrator is the brain that ties it all together. It is the planner, the coder, and the delegator\u2014all in one model.<\/p>\n\n\n\n
Its job is to take a messy request\u2014like \u201cBook me a dentist appointment Tuesday afternoon and add it to my calendar\u201d\u2014and convert it into a concrete plan.<\/p>\n\n\n\n
The Orchestrator figures out the steps, picks the right tool or sub-agent for each step, writes code when needed, and recovers when something breaks mid-task.<\/p>\n\n\n\n
What\u2019s interesting is that the recipe is quite simple. Orchestration is usually where people reach for the biggest model they can get, but we wanted to show you that you can push all of this into a small model without giving up capability.<\/p>\n\n\n\n
The training is standard SFT, but the key is the data mix\u2014blending complementary styles of data in the right ratio.<\/p>\n\n\n\n
The first style is tool-calling data<\/strong>\u2014clear requests, selecting tools, calling them with arguments, and handling responses.<\/p>\n\n\n\nThe second is terminal-style data<\/strong>, where an agent performs step-by-step actions\u2014observing results and deciding what to do next.<\/p>\n\n\n\nMixing these teaches the model when to use tools and when to generate code directly.<\/p>\n\n\n\n
The result is a model that competes with much larger ones, while staying small enough to run locally alongside Fara\u2014and it\u2019s open weight, so it can integrate into your own systems.<\/p>\n\n\n\n
With that, Hussein will introduce Fara.<\/p>\n\n\n\n
HUSSEIN MOZANNAR:<\/strong> One of our goals in AI Frontiers is to train agentic models to complete computer-use tasks. Our bet is that end-to-end synthetic data generation\u2014without human interaction data\u2014can get us there.<\/p>\n\n\n\nLast November, we released Fara-7B. Today, we\u2019re excited to introduce Fara-1.5<\/strong>, a family of models across three sizes: 4B, 9B, and 27B.<\/p>\n\n\n\nFara-1.5 sets state-of-the-art results for models in its class.<\/p>\n\n\n\n
Fara can handle real-world web tasks like form filling, booking, shopping, and other repetitive actions. It works by capturing screenshots, analyzing them alongside past context, and predicting the next action\u2014like clicking or typing.<\/p>\n\n\n\n
It operates in a loop: observe, act, evaluate, and continue.<\/p>\n\n\n\n
Its action space includes clicking, keyboard input, memory tools, and the ability to ask the user for approval when needed.<\/p>\n\n\n\n
On the Mind2Web benchmark, Fara-1.5 nearly doubles performance compared to Fara-7B, improving from 35% to about 65%.<\/p>\n\n\n\n
But we didn\u2019t train it just for benchmarks\u2014we trained it for real-world usefulness.<\/p>\n\n\n\n
This is enabled by our synthetic data system, FaraGen 2.0<\/strong>, which generates training data using:<\/p>\n\n\n\n\n- Live and synthetic web environments<\/li>\n\n\n\n
- A strong teacher agent<\/li>\n\n\n\n
- A user simulator<\/li>\n\n\n\n
- Verification systems for correctness, efficiency, and safety<\/li>\n<\/ul>\n\n\n\n
This allows us to scale data generation and train models effectively.<\/p>\n\n\n\n
Looking ahead, we plan to expand Fara with always-on capabilities, support for additional environments like Windows and Linux, and deeper integration with terminal workflows.<\/p>\n\n\t\t\t\t<\/span>\n\t\t\t<\/div>\n\t\t\t\n\t\t\t\tShow more\t\t\t<\/button>\n\t\t<\/div>\n\t<\/div>\n<\/div>\n","protected":false},"excerpt":{"rendered":"What if you could run a capable AI agent without leaning on frontier-scale models? MagenticLite is the next generation of Magentic-UI, an agentic experience reimagined and optimized for small language models. It works across both your browser and your local file system in a single workflow, keeping you in the driver’s seat at every step. […]<\/p>\n","protected":false},"featured_media":1171928,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","msr_hide_image_in_river":0,"footnotes":""},"research-area":[13556],"msr-video-type":[268311],"msr-locale":[268875],"msr-post-option":[],"msr-session-type":[256174],"msr-impact-theme":[],"msr-pillar":[],"msr-episode":[270329],"msr-research-theme":[270110],"class_list":["post-1171587","msr-video","type-msr-video","status-publish","has-post-thumbnail","hentry","msr-research-area-artificial-intelligence","msr-video-type-microsoft-research-forum","msr-locale-en_us"],"msr_download_urls":"","msr_external_url":"https:\/\/youtu.be\/wKMQ3CZig_8","msr_secondary_video_url":"","msr_video_file":"http:\/\/0","_links":{"self":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-video\/1171587","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-video"}],"about":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-video"}],"version-history":[{"count":9,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-video\/1171587\/revisions"}],"predecessor-version":[{"id":1172101,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-video\/1171587\/revisions\/1172101"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/media\/1171928"}],"wp:attachment":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1171587"}],"wp:term":[{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1171587"},{"taxonomy":"msr-video-type","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-video-type?post=1171587"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1171587"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=1171587"},{"taxonomy":"msr-session-type","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-session-type?post=1171587"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=1171587"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=1171587"},{"taxonomy":"msr-episode","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-episode?post=1171587"},{"taxonomy":"msr-research-theme","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-theme?post=1171587"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}