How work moves through Rovai

Understand who decides, who executes, and where the files and results live.

Responsibilities and data flow

You choose a goal and recipients. Rovai routes work through a teammate’s configured Agent. That Agent may contact its configured model service and operate on the host’s project files.
You choose a goal and recipients. Rovai routes work through a teammate’s configured Agent. That Agent may contact its configured model service and operate on the host’s project files. Mermaid source

Five different roles

You
Set the goal, choose recipients, inspect output and decide whether to approve an operation or accept a delivery.
Rovai
Organize teammates and conversations, queue addressed messages, launch configured Agents and retain observable execution facts.
Teammate
A lasting identity with responsibilities and configuration. Several teammates can use the same Agent product with different roles.
Native Agent
Products such as Codex CLI or Claude Code interpret the request, call the configured model and use tools under their own capabilities and permissions.
Project and services
Files are on the execution host. Model requests follow the chosen Agent/provider configuration; a local workspace is not a guarantee that all inputs stay on the computer.

Follow one request

  1. Select Dingding in the mention picker and send a scoped request in the Orbit conversation.
  2. Rovai accepts the message and records a delivery for that teammate. When execution can start, it freezes the selected input and configuration into a Run.
  3. The Agent reads or edits files and returns observable steps and output. Read the public answer, the Run and the actual files together.
  4. To request review, select Cheese and send a new request. That creates new work; merely naming a reviewer in prose is not the same action.

What is shared, and what is separate

Participants in a project conversation work against its selected workspace. A private one-on-one has separate messages and a separate queue, but it does not automatically create an isolated file copy.

A Git-backed Mission can have a shared managed worktree. Its teammates use that Mission execution location together; each teammate does not automatically get a worktree.

Permissions and actual model support come from the current Agent configuration. A profile description, a Skill or an @ mention does not grant filesystem or network access.