Need smarter insights in your inbox? Join our weekly newsletters to get solely what issues to enterprise AI, knowledge, and safety leaders. Subscribe Now
Liquid AI, the startup shaped by former Massachusetts Institute of Expertise (MIT) researchers to develop novel AI mannequin architectures past the widely-used “Transformers“, immediately introduced the discharge of LEAP aka the “Liquid Edge AI Platform,” a cross-platform software program improvement equipment (SDK) designed to make it simpler for builders to combine small language fashions (SLMs) immediately into cell purposes.
Alongside LEAP, the corporate additionally launched Apollo, a companion iOS app for testing these fashions regionally, furthering Liquid AI’s mission to allow privacy-preserving, environment friendly AI on client {hardware}.
The LEAP SDK arrives at a time when many builders are searching for options to cloud-only AI companies attributable to issues over latency, value, privateness, and offline availability.
LEAP addresses these wants head-on with a local-first method that permits small fashions to run immediately on-device, decreasing dependence on cloud infrastructure.
The AI Impression Collection Returns to San Francisco â August 5
The following section of AI is right here â are you prepared? Be a part of leaders from Block, GSK, and SAP for an unique have a look at how autonomous brokers are reshaping enterprise workflows â from real-time decision-making to end-to-end automation.
Safe your spot now â house is restricted: https://bit.ly/3GuuPLF
Constructed for cell devs with no prior ML expertise required
LEAP is designed for builders who need to construct with AI however might not have deep experience in machine studying.
Based on Liquid AI, the SDK could be added to an iOS or Android challenge with just some strains of code, and calling a neighborhood mannequin is supposed to really feel as acquainted as interacting with a standard cloud API.
“Our analysis reveals builders are shifting past cloud-only AI and in search of trusted companions to assist them construct on-device,” stated Ramin Hasani, co-founder and CEO of Liquid AI, in a weblog submit saying the information immediately on Liquid’s web site. “LEAP is our reply—a versatile, end-to-end deployment platform constructed from the bottom as much as make highly effective, environment friendly, and personal edge AI actually accessible.”
As soon as built-in, builders can choose a mannequin from the built-in LEAP mannequin library, which incorporates compact fashions as small as 300MB — light-weight sufficient for contemporary telephones with as little as 4GB of RAM (!!) and up.
The SDK handles native inference, reminiscence optimization, and gadget compatibility, simplifying the everyday edge deployment course of.
LEAP is OS- and model-agnostic by design. At launch, it helps each iOS and Android, and presents compatibility with Liquid AI’s personal Liquid Basis Fashions (LFMs) in addition to many well-liked open-source small fashions.
The aim: a unified ecosystem for edge AI
Past mannequin execution, LEAP positions itself as an all-in-one platform for locating, adapting, testing, and deploying SLMs for edge use.
Builders can browse a curated mannequin catalog with numerous quantization and checkpoint choices, permitting them to tailor efficiency and reminiscence footprint to the constraints of the goal gadget.
Liquid AI emphasizes that enormous fashions are typically generalists, whereas small fashions usually carry out finest when optimized for a slim set of duties. LEAP’s unified system is structured round that precept, providing instruments for speedy iteration and deployment in real-world cell environments.
The SDK additionally comes with a developer group hosted on Discord, the place Liquid AI presents workplace hours, help, occasions, and competitions to encourage experimentation and suggestions.
Apollo: like Testflight for native AI fashions
To enhance LEAP, Liquid AI additionally launched Apollo, a free iOS app that lets builders and customers work together with LEAP-compatible fashions in a neighborhood, offline setting.
Initially a separate cell app startup that allowed customers to talk with LLMs privately on gadget, which Liquid acquired earlier this yr, Apollo has been rebuilt to help the whole LEAP mannequin library.
Apollo is designed for low-friction experimentation — builders can “vibe verify” a mannequin’s tone, latency, or output habits proper on their telephones earlier than integrating it right into a manufacturing app. The app runs completely offline, preserving consumer privateness and decreasing reliance on cloud compute.
Whether or not used as a light-weight dev instrument or a personal AI assistant, Apollo displays Liquid AI’s broader push to decentralize AI entry and execution.
Constructed on the again of the LFM2 mannequin household introduced final week
LEAP SDK launch builds on Liquid AI’s July 10 announcement of LFM2, its second-generation basis mannequin household designed particularly for on-device workloads.
LFM2 fashions are available in 350M, 700M, and 1.2B parameter sizes, and benchmark competitively with bigger fashions in velocity and accuracy throughout a number of analysis duties.
These fashions kind the spine of the LEAP mannequin library and are optimized for quick inference on CPUs, GPUs, and NPUs.
Free and prepared for devs to start out constructing
LEAP is at present free to make use of underneath a developer license that features the core SDK and mannequin library.
Liquid AI notes that premium enterprise options shall be made accessible underneath a separate business license sooner or later, however that it’s taking inquiries from enterprise prospects already by way of its web site contact kind.
LFM2 fashions are additionally free for educational use and business use by corporations with underneath $10 million in income, with bigger organizations required to contact the corporate for licensing.
Builders can get began by visiting the LEAP SDK web site, downloading Apollo from the App Retailer, or becoming a member of the Liquid AI developer group on Discord.
