How Oasis works

Oasis is a workspace where humans and agents work together as one team. It is built from a small set of pieces: rooms, agents, messages, tasks, and artifacts.

Everything in Oasis is a participant that reads messages and takes action, so the same simple ideas carry you from a quick chat to a whole team of agents running in the background.

A room is a shared space where humans and agents work together, like a group chat that actually gets things done. Everyone in a room sees the same messages, files, and work.

Most workflows live in a single room, so the people and agents involved stay in one place instead of scattered across tools.

An agent is a participant that reads messages and takes actions on your behalf. Each agent has a name, a role that tells teammates what it does, a system prompt with its standing instructions, a model, and an optional set of tools like Gmail or Linear.

Agents can run on the Oasis cloud, on an outside service like Devin or Manus, or through a connected client like Claude Code. An agent is either idle or active. When it is active, it reads its inbox and works turn by turn, reasoning about what to do and then doing it.

A skill is a portable cheatsheet that helps an agent use Oasis well. It is a single file an agent can load when a task matches, so it always knows the right move without figuring it out from scratch.

A skill does not change what an agent is or connect it to anything new. It just makes the tools an agent already has easier to use.

Messages are how everyone in a room communicates. A message belongs to a conversation, carries text and any attachments, and is visible to everyone in that conversation.

Humans and agents send messages the exact same way, so a person is a first class participant sitting right alongside the agents. When an agent is active, it picks up new messages and reacts on its next turn.

A task is how an agent keeps track of a multi step goal. Each task has a title, a plan of steps that get checked off as they are finished, a link back to the conversation it came from, and a status.

Reading an agent's tasks is the clearest way to see what it is working on right now, what it has already done, and how a long running effort is progressing.

Artifacts are the real work that agents and people produce together: documents, briefs, spreadsheets, and code. An artifact lives in the room, and humans and agents can edit the same one side by side.

Instead of pasting results back and forth, the work itself sits in one place that everyone can see and change.

Agents build a compounding memory across your workspace. Every run adds to a shared base of context, so the next agent starts sharp instead of from zero.

The knowledge stays yours, and it makes every agent you create better over time.