AI Audio2024
Runs text-to-speech, speech-to-text, and speech-to-speech models natively on Apple Silicon via MLX — no CUDA or cloud. Supports 20+ TTS and 15+ STT models (Kokoro, Whisper, Qwen3), low-bit quantization, an OpenAI-compatible API, and a Swift package.
AI Infra2024
Connects AI agents to 50+ apps and databases — Notion, Slack, Salesforce, GitHub, Jira — then continuously syncs and indexes their data behind one search API, with auth, ingestion, and retrieval exposed via MCP, REST, and SDKs.
AI Train2024
A 671B-parameter Mixture-of-Experts language model (37B activated) trained on 14.8T tokens with 128K context, FP8-first training, a Multi-Token Prediction module, and Hugging Face weights—focused on efficient MoE training and long-context use cases.
Embodied AI2024
Provides an open platform of omnimodal world models, datasets, and tools to build Physical AI — joint perception, generation, and action reasoning for robots, autonomous vehicles, and smart infrastructure. Supports images, video, audio, and action-conditioned workflows.
AI Agent2025
Lets teams build, deploy, and manage AI agents from chat, visual workflows, code, knowledge bases, tables, and more than a thousand integrations.
AI Agent2025
Runs penetration tests autonomously: a multi-agent system (researcher, developer, executor) plans attacks, writes and runs exploit code, and chains 20+ tools like nmap, metasploit and sqlmap in isolated Docker containers — for authorized testing only.
AI Agent2025
Runs stateful AI agents as Cloudflare Durable Objects — each keeps its own storage and lifecycle, hibernating when idle and waking on demand. Adds WebSocket state sync, type-safe RPC, resumable LLM streaming, MCP roles, and durable workflows.
AI Deploy2025
Provides a hardware plugin that runs vLLM on Huawei Ascend NPUs by mapping vLLM execution and memory management to the Ascend runtime. Key features: support for Transformer/MoE/embedding/multimodal models, official docs, CI-backed release branches and community maintenance.
AI Infra2025
High-performance CUDA tensor-core GEMM kernel library for LLM workloads: supports FP8/FP4/BF16, fused Mega MoE and MQA scoring, and runtime JIT-compiled kernels. Targets NVIDIA SM90/SM100 and PyTorch—for teams working on low-level GPU kernel optimization.
AI Infra2025
Provides high-throughput, low-latency GPU communication kernels for Mixture-of-Experts (MoE) and expert-parallel workloads, with NVLink↔RDMA-aware forwarding, FP8/BF16 support, and low-latency RDMA hooks for inference decoding.
AI Infra2025
Optimized MLA (Multi-head Latent Attention) decoding kernels powering DeepSeek-V3/V3.2 inference on Hopper and Blackwell GPUs. Dense decoding reaches ~3000 GB/s and 660 TFLOPS on H800; the sparse path stores the KV cache in FP8.