Rysh and Claude Code subagents: how they fit together
Claude Code can already run many agents: subagents, agent teams and background sessions are in its own docs. Rysh does not replace any of that. It is the workspace around it: each Claude Code (or other agent) gets its own pane and its own git worktree, and the whole session lives in a daemon, so it keeps running when you close the terminal.
What each one is
Claude Code (as of 13 November 2026, Anthropic’s docs): “Each subagent runs in its own context window with a custom system prompt, specific tool access, and independent permissions.” A subagent “with isolation: worktree runs its Bash and PowerShell commands inside its worktree.” Agent teams “coordinate multiple Claude Code instances working together as a team, with shared tasks, inter-agent messaging”, and Agent view “is one screen for all your background sessions”.
Rysh: Claude Code runs unchanged, full-screen, inside a rysh pane. Rysh can spawn a fleet of agents from files on disk, each in its own git worktree, so they cannot overwrite each other. Detach and only the terminal UI closes: the session daemon keeps running, so an agent’s task finishes with no UI attached. The difference is where you watch the work: inside one conversation, or one pane per agent.
Feature comparison
| Feature | Rysh | Claude Code on its own (as of 13 November 2026) |
|---|---|---|
| Several agents at once | ✓ one pane per agent, prompted from one input box | ✓ subagents, agent teams, background sessions |
| Each agent in its own git worktree | ✓ isolation: worktree in the agent file | ✓ isolation: worktree for subagents |
| Runs Claude Code itself | ✓ full-screen in a pane | (is Claude Code) |
| Other agent CLIs in the same workspace | ✓ e.g. Codex CLI, aider panes | ✗ Claude only |
| Work keeps running after you close the terminal | ✓ session daemon; rysh attach to come back | ✓ background sessions |
| Share a live pane with a teammate on another machine | ~ per-share shell limits; no hosted share link | ✓ cloud sessions / Remote Control (their docs) |
| Policy in front of agent tool calls | ✓ seven rule classes, org policy strictest-wins | ✓ permission modes, hooks |
| Risky tool calls stop for approval in the pane | ~ pane agents yes; spawned agents auto-approve by default | ✓ permission modes |
| Bring your Claude Code settings / MCP config across | ✓ rysh import (preview first, nothing written without --apply) | n/a |
| OS sandbox for shell commands | ~ opt-in ##claude --sandbox pane only | ✓ Seatbelt / bubblewrap |
| Checkpoints / rewind, IDE extensions, native Windows | ✗ | ✓ |
Claude Code facts are from Anthropic’s own docs, as of 13 November 2026.
Which should you choose?
Choose Rysh if…
- You run several agents at once and want to see each one in its own pane.
- You mix Claude Code with other agent CLIs.
- You want the session to survive closing the terminal on your own machine.
- You want one policy file in front of every agent.
Choose Claude Code if…
- You work in one repo, one conversation at a time.
- You want
/rewindcheckpoints, the IDE extensions or native Windows. - You want its cloud sessions on a phone.
FAQ
Does rysh replace Claude Code?
No. Claude Code runs unchanged inside a rysh pane. Rysh runs on your model key; plan prices exclude model usage.
Can I keep using subagents inside rysh?
Yes: a Claude Code pane in rysh is ordinary Claude Code.
Where do my agents run?
Today rysh runs your agents on your own machine (macOS and Linux). Hosted agents, where your agents run on our servers and keep working when your laptop is closed, are being built. They are not available yet. To hear when they are, write to info@rysh.ai.
What does it cost?
There is a free plan; see rysh.ai/pricing.
Try Rysh
Share a pane, a lane or a group with a teammate on another machine — see it in short videos or the written guides.
Download rysh →