Projects
A project groups the conversations that belong to the same body of work, and carries context every conversation inside it can read.
When to use it
Use a project when you notice yourself re-explaining the same background in thread after thread — the client, the codebase, the launch, the quarter. The project holds that background once.
| Put it in | When |
|---|---|
| The system prompt | It is true of everything the agent ever does |
| A project | It is true of this body of work |
| Memory | It is true of you, across all work |
| A message | It is true of this one task |
What a project gives an agent
A conversation inside a project starts with the project’s context available. The agent can also read the project’s current state and update it as work progresses — so a status the agent maintains stays current without you transcribing it.
That makes a project useful as a shared working record: what this work is, where it stands, what was decided.
Working in one
Open the project and start a conversation from it, and that conversation belongs to the project. The project page lists its conversations, so returning to a body of work does not mean hunting through a global recency list.
Limits
- A conversation belongs to one project.
- Project context is available to conversations in that project — it is not shared with the agent’s other threads.
- Projects group work; they do not grant access, and they are not a place to store documents. Results belong in outputs.