Kordi: Building AI-Native Supercollaboration

AI assistants are usually designed as private, one-person tools. Collaboration is not. Real work happens in shared conversations—with teammates, decisions, context, and different agents that can help at the right moment.

Kordi treats AI as a participant in the conversation. People can chat one-to-one or in groups, bring personal agents into the same shared space, and keep the experience synchronized through native desktop and iPhone apps.

People and agents, together

Invite an agent into the conversation instead of moving the conversation into a separate AI tool.

Familiar social primitives

Accounts, contacts, direct chats, groups, unread state, and synchronized history.

Agent-native groups

Mention your own Kordi agent—or another participant’s agent—inside the group where the context already lives.

Hosted continuity

A hosted API and agent runner keep collaboration and execution available beyond one local process.

Kordi is under active development.
Visit the GitHub
Our results reveal systematic gaps: frontier LLMs frequently fail to maintain stable prioritization under conflicting user objectives, exhibit increasing privacy violations over multi-turn interactions, and suffer from efficiency bottlenecks when coordination requires iterative information gathering.”

Multi-User Large Language Model Agents

Shu Yang, Shenzhe Zhu, Hao Zhu, José Ramón Enríquez, Di Wang, Alex Pentland, Michiel A. Bakker, and Jiaxin Pei

Read publication