AI Deploy2025
Run large-language and multimodal models locally on edge devices (Android, iOS, desktop, web, Raspberry Pi) with hardware acceleration, function-calling, and multi-language SDKs—designed for low-latency, privacy-sensitive on-device inference.
AI Agent2025
Framework for building an organization's internal coding agents — runs tasks in isolated cloud sandboxes, integrates with Slack/Linear/GitHub, orchestrates subagents, and automates commits/PRs. Built on LangGraph and Deep Agents for easy customization.
AI Infra2025
Extends RAG beyond text: parses PDFs and Office files containing images, tables, equations, and charts, then queries them through one multimodal knowledge graph. Built on LightRAG, it replaces separate parsing and retrieval tools.
AI Infra2025
Reimplements the vLLM inference engine from scratch in ~1,200 lines of readable Python, matching its offline throughput on small models. Prefix caching, tensor parallelism, torch.compile, and CUDA graphs are all kept legible.
AI Train2025
Trains and optimizes AI agents with reinforcement learning using almost zero code change. Works with any agent framework (LangChain, OpenAI Agents SDK, AutoGen, CrewAI) or none, and can selectively optimize a single agent inside a multi-agent system.
AI Agent2025
Transforms Claude Code into a structured development platform by injecting behavioral instructions and orchestrating workflows via 30 slash commands. Provides 20 specialized agents and optional MCP server integrations for faster, token‑efficient research and agent-driven dev workflows.
AI Agent2025
Chinese-enhanced fork of TradingAgents that runs multi-agent LLM stock analysis for A-share, HK and US markets, adding domestic models (Qwen, DeepSeek) and local data sources (Tushare, AkShare, BaoStock), with report export to Word and PDF.
AI Agent2025
Evaluates and optimizes AI agents and language models in containerized environments, supporting large-scale parallel benchmarks and RL rollouts. Integrates with third‑party providers for thousands of parallel environments and serves as the official harness for Terminal‑Bench.
AI Agent2025
Deep research agent for complex, long-horizon research and prediction tasks. Pairs a 256K context window with up to 300 tool calls per query for web search, extraction, and code execution. Ships as open 30B and 235B models scoring 82.7% on GAIA.
MCP Server2025
Provides an MCP server exposing 30+ trading tools — real-time prices, technical indicators, Bollinger Band scores, Reddit/news sentiment, and backtesting — designed to integrate with Claude/OpenClaw agents for automated market analysis.
AI Deploy2025
A ~5,000-line Python LLM inference engine that re-implements SGLang's serving optimizations — radix KV-cache reuse, chunked prefill, overlap scheduling, tensor parallelism — as a fully type-annotated reference instead of a black box.