Create agent systems.
Go from using AI to having AI systems running.
Build organizations of agents that work together, from one agent to hundreds.
What would you put in motion?
- Legal
A legal desk that checks every citation.
Researchers pull the authorities, drafters write the memo, and a cite checker verifies each quote before a client sees it.
- Growth
An online business that runs its own experiments.
Researchers find customers, writers build the pages, developers ship, and an analyst reports what moved the numbers.
- Content
A content studio that researches before it writes.
Sources, drafts and reviews collect in each piece's thread, and nothing is scheduled until it's approved.
- Fundraising
A fundraise with a thread for every investor.
Researchers study each firm, editors draft the introductions, and every message waits for approval.
The building blocks of an agent system.
- Agents
Agents are software systems.
Each agent is a persistent software system with its own applications and behavior. Conversation is its interface.
- Components
A modular agent harness.
Plug-and-play mini-apps extend the harness itself, adding their own state, interfaces and runtime behavior.
- Conversations
Programmable threads.
Threads define how agents collaborate on a unit of work, what their contributions trigger and how the work converges.
- Workflows
Multi-agent orchestration.
A workflow orchestrates multiple agents in its own working context, then returns the result to the ongoing conversation.
- Coding agents
Agent-native.
Codex and Claude Code can work inside a persistent agent system and build and operate it from outside, using the tools you already know.
Runs on your computer.
Agents and their data stay on your machine. Sign in with a ChatGPT account to start.
Built in code.
Each system is a short file. Save a change and the running system updates right away.