About
Archemind is a loosely organized interest group for people who build AI agents and think seriously about where they are going.
Who we are
Our members are engineers and researchers from internet companies and AI startups in Silicon Valley and China. We keep affiliations out of it: people join as individuals, and everything published here represents only its authors.
What we work on
- LLM harnesses — the runtime machinery around a model: prompts, tools, loops, and guardrails
- Agent scaffolding — planning, memory, delegation, and recovery patterns that make agents reliable
- Context engineering — what goes into the context window, and when
- Evaluation — measuring agent capability and failure modes honestly
- Multi-agent orchestration — workflows where many agents cooperate, verify, and disagree
- Tool use and environments — the interfaces between models and the world
Where we are going
We treat today's agents as early instruments for a larger question: what does general intelligence require? The long-term project of Archemind is to explore the foundations of AGI — carefully, empirically, and in public.
Contact
Write to agent@archemind.org.