Haiyang Li (ohdearquant)
Haiyang (Ocean) Li is an individual builder and solo developer behind an open-source AI-agent software stack. His stated focus is memory and coordination infrastructure for long-horizon agents (khive), agent orchestration (LionAGI), and local inference (Lattice), with projects designed to work together.
At a Glance
- Developers and engineering teams building LLM applications
- Teams operating coding agents and multi-agent workflows
- Organizations needing long-horizon agent memory and coordination
- Researchers and developers experimenting with local inference
- +1 more
AI Tools by Haiyang Li (ohdearquant)
(1)
lionagi
Python Multi Agent Orchestration
Discussions
No discussions yet
Be the first to start a discussion about Haiyang Li (ohdearquant)
Latest News
LionAGI 0.35.2 released on PyPI; Haiyang Li listed as author and ohdearquant as maintainer.
Introducing khive, an open-source Rust knowledge-graph runtime for AI agents and single-binary MCP server.
Agentics NZ hosted a special session with Li on Lion AGI; the listing identifies him as creator of Lion AGI and co-founder of Agentics Foundation.
Agentics NZ published recordings and highlighted Li presenting LionAGI, khive.ai, and the Lattice inference engine from New York.
Products & Services
Apache-2.0 open-source, governed multi-agent orchestration framework for Python, with typed and inspectable state, model/provider integration, structured output, parallel fan-outs, dependency-aware DAG flows, durable runs, and the li CLI.
Apache-2.0 open-source Rust knowledge-graph and memory runtime for AI agents. It provides typed entities, a closed edge ontology, hybrid lexical/vector search, durable task state, and MCP access through the kkernel single binary.
Pure-Rust local inference engine focused on Apple Silicon and Linux, including transformer inference, embeddings, quantization, LoRA fine-tuning, and a native macOS app/instrument panel.
Local-first orchestration and observability studio built on LionAGI. It provides live agent runs, schedules, execution graphs, artifacts, telemetry, and run history while talking to a local li daemon.
Market Position
Li's stack positions itself as an open-source, inspectable and controllable alternative to higher-level agent frameworks: LionAGI's documentation explicitly provides a comparison with LangChain, LangGraph, LlamaIndex, and AG2. Its differentiation is durable local runs, governance controls, CLI/API parity, and a vertically integrated memory–orchestration–inference stack rather than a hosted-only abstraction.
Leadership
Founders
Haiyang (Ocean) Li
Founder of khive AI and creator of LionAGI; co-founder of Agentics Foundation and AG2 maintainer. His public GitHub profile describes him as building agent memory and coordination infrastructure in Rust and Python and lists New York, NY.
Executive Team
Haiyang (Ocean) Li
Founder, khive AI; creator of LionAGI and Lattice; co-founder, Agentics Foundation
Open-source AI infrastructure builder and AG2 maintainer. His GitHub profile says he designs the architecture and constraints for the agent-built stack; public event material identifies him as the creator of Lion AGI and co-founder of Agentics Foundation.
Founding Story
Li's software work began with LionAGI in 2023, described in the repository citation as 'LionAGI: Towards Automated General Intelligence.' The work has since expanded into a vertically integrated open-source stack: khive for memory and coordination, LionAGI for orchestration, Lattice for inference, and LNkernel for formal verification.
Business Model
Revenue Model
The software projects are open source and self-hostable. The associated khive offering also promotes a managed cloud endpoint/waitlist, while Li offers prepaid AI architecture and agent-systems consulting sessions.
Pricing Tiers
Architecture gut-check, build-vs-buy decision, and quick sanity check.
System design review, technical due diligence, implementation planning, and written follow-up.
End-to-end system design, code review, agent orchestration planning, written brief, and 30-day async follow-up.
Target Markets
- Developers and engineering teams building LLM applications
- Teams operating coding agents and multi-agent workflows
- Organizations needing long-horizon agent memory and coordination
- Researchers and developers experimenting with local inference
- Companies seeking production agent architecture, technical due diligence, or implementation planning
- Building and running multi-step LLM workflows
- Coding-agent workflows such as code review and security audits
- Research synthesis and parallel specialist work
- Long-running, resumable, scheduled, and monitored agent operations
- Long-horizon agents that need durable memory, knowledge graphs, tasks, and coordination
- Local inference and model experimentation on Apple Silicon or Linux