Liquid AI Integrations

The LEAP Edge AI Platform offers a full-stack on-device AI toolchain that enables developers to build edge AI applications, from model selection through inference, entirely on device. It includes a best-model search engine to find the most appropriate model for a given task and device constraint, a curated library of pre-trained model bundles ready for download, and fine-tuning tools (such as GPU-optimized scripts) for customizing models like LFM2 to specific use cases. It supports vision-enabled capabilities across iOS, Android, and laptop devices, and includes function-calling so AI models can interact with external systems via structured outputs. For deployment, LEAP provides an Edge SDK that lets developers load and query models locally, just like a cloud API, but entirely offline, and a model bundling service to package any supported model or checkpoint into a bundle optimized for edge deployment.

Apollo is a lightweight mobile application designed for fully on-device, cloud-free AI interactions, enabling users to engage with advanced language and vision models securely, privately, and with low latency. It supports a library of small foundation models from the company’s LEAP platform, allowing users to draft messages, emails, chat with a private AI assistant, craft digital characters, or use image-to-text capabilities, all without an internet connection and with no data leaving the device. Apollo is optimized for real-time responsiveness and offline operation, ensuring that inference happens entirely locally, with no API calls, servers, or user-data logging involved. It serves as both a personal AI playground and a testing bed for developers using LEAP models, letting one “vibe-check” how a model performs on their own mobile hardware before broader deployment.

SF Compute is a marketplace platform that offers on-demand access to large-scale GPU clusters, letting users rent powerful compute resources by the hour, not requiring long-term contracts or heavy upfront commitments. You can choose between virtual machine nodes or Kubernetes clusters (with InfiniBand support for high-speed interconnects), and specify the number of GPUs, duration, and start time as needed. It supports flexible “buy blocks” of compute; for example, you might request 256 NVIDIA H100 GPUs for three days at a capped hourly rate, or scale down/up dynamically depending on budget. For Kubernetes clusters, spin-up times are fast (about 0.5 seconds); VMs take around 5 minutes. Storage is robust, including 1.5+ TB NVMe and 1 TB + RAM, and there are no data transfer (ingress/egress) fees, so you don’t pay to move data. SF Compute’s architecture abstracts physical infrastructure behind a real-time spot-market and dynamic scheduler.

LFM-40B offers a new balance between model size and output quality. It leverages 12B activated parameters at use. Its performance is comparable to models larger than itself, while its MoE architecture enables higher throughput and deployment on more cost-effective hardware.