Docs / Agents
Agents and shared sessions
Start, share, steer and continue agent work across runtimes and computers.
An agent session holds a task's conversation and tool activity. Start one for each job, and name the repository or working folder in the first message. Choose New agent, then pick an online runtime in the composer. A local path must exist on that runtime's computer.
Select tools and credentials
Select the connections the task needs. A selected team connection is still limited by the member allow-list and the provider's permissions. Personal credentials belong to their member. Inviting someone into a session does not give them your personal accounts. Some other entry points select credentials differently, so check their setup guide before you start work there.
Mention a connected resource in the prompt when you want the agent to inspect that exact account, cluster, deployment or issue. Resource context helps resolve names. It never grants a credential you didn't select.
Share a session
Use the session's Share control to review who has access. You can also @mention a teammate and choose Invite. Check whether you are giving read access or permission to contribute. Participants see the same transcript, and can steer the work only when their session and computer permissions allow it.
Sharing the conversation and sharing its computer are separate. A teammate who can't use the machine may still read the session without sending work there. Read computer sharing before you invite someone to a local runtime: its sign-ins are powerful.
To end access, remove the participant or change the session's general access in its sharing controls. Removing someone from one conversation does not revoke their access to the computer or a connector.
Steer, answer and stop
Tool calls and results appear in the transcript. Send a message during a turn to steer it. Answer question cards in Needs your input or the Inbox. Stop the running turn when you need it to halt. Stopping does not undo commands that already ran.
Permission modes decide which actions ask first. Plans cover significant cloud infrastructure work, not every code edit or ticket. Say so when you want to review a diff, commit or publication before it happens.
Move work to another runtime
Use the session's runtime picker and read the move confirmation. Prepare the repository, branch, dependencies and sign-ins on the new computer. A different computer does not have the files from the old one.
Keep changes in git or copy the files you need, then ask the agent to check its working folder before it continues. History gives task context. A move may continue in a new provider session rather than the original one.
Continue an existing CLI session
For local runtimes, start a new agent, choose +, then Continue a Claude Code session or Continue a Codex session. Pick a session from that computer and review the import. Long imports may leave out older turns. Importing history does not clone the repository.
Search and return later
Use search (⌘K) to find earlier conversations. Archive finished agents to reduce clutter. Find archived work under Archive at the foot of the sidebar's chat list, or in Settings → Archived agents. An archived transcript is not the same as a deleted computer.
History is saved, and long context may be compacted. A cloud computer or your own server can keep working after you close your laptop, but a local runtime needs its computer and Pangea running. A saved transcript does not mean a background process survived a restart.
Other entry points
Slack and other connected chat channels can continue agent work in their own threads. Scheduled and webhook work uses triggers. Their starting permission mode and credential selection can differ from a session in the app, so check those settings before you point them at production.