Docs / Get started
Core concepts
Teams, shared sessions, runtimes, computers, and the limits around their tools and context.
These concepts describe who takes part in the work, where it runs, and which context and access it can use.
Teams and sessions
A team groups members, connected tools and shared settings. Each member is an Administrator, an Editor or a Viewer. Administrators manage members, billing and team connectors. Editors work with agents and the connections they are allowed to use. Viewers have read-only access. See Teams and members.
An agent session is the conversation you give a job to, with its transcript and task context. The app also calls it an agent. Start a separate session for each job. Sharing a session lets teammates read or help, based on its access settings. Team membership alone does not open every private conversation, or the computer that runs it.
Computers and runtimes
- A computer is a machine: your Mac, your own server, or a Pangea Cloud computer.
- A runtime is a coding agent installed there, such as Claude Code, Codex, Cursor or Antigravity. It keeps its own harness and its provider sign-in.
- A session runs on a runtime. Moving it changes where future work runs. It does not copy repositories, secrets or local files to the new machine.
Supported runtimes differ in sign-in, models and platform support. Local work needs your Mac to stay on. Remote work keeps going after you disconnect. Cloud computers come with your plan. Model usage is billed by your model provider, never by Pangea.
Workspace and context
The Mac app can show several sessions in split panes, with tools in the dock beside them: browser tabs, terminals, desktops and artifacts. Connector pages let you inspect accounts and resources and start an agent with that context. They are convenient views, not the limit of what a provider's CLI or API can do. See Desktop app.
Shared and personal knowledge
Instructions give standing guidance. Skills package procedures and supporting files. Memory stores durable facts and verified playbooks to look up later. Team and personal scopes are separate. Sharing a team does not share your personal settings or credentials. Agents save and search memory with tools; they don't get the whole memory store in every prompt. See Instructions, Skills and Memory.
Credentials and permissions
Connect the tools a task needs, then select which connections the session may use. Team roles, connector allow-lists, the session's selection and the provider identity's own permissions all apply. Personal connections, such as GitHub, Linear, Gmail or Drive, belong to the member who connected them.
Federated cloud connections mint short-lived credentials. Other connectors store an encrypted token or client secret, or proxy access on the server. A local runtime can also use sign-ins already on its computer, which is a reason to share machines with care. See Access control and IAM trust model.
Plans and approvals
A plan makes significant cloud infrastructure work reviewable: scope, decisions, steps, risk and cost. It is not a gate on every write. Auto Mode and Full Access control the runtime's permission prompts for single actions. Neither can grant a cloud permission the connected identity lacks. See Plans and permission modes.
Automation
A trigger starts work on a schedule or from a webhook. A Watch wires a provider alert to an agent workflow. Both are saved setups. A command left running in a session is not a durable monitor. Review the runtime, credentials, permission mode and destination before you automate a task. See Triggers and automation.