Every agent gets its own live terminal, tiled across panes, tabs and windows, all pointed at the same Unity project, with the game they are changing running beside them.
Detected from your PATH. Bring the subscription you already pay for.
A single terminal makes agent work serial: ask, wait, read, verify, ask again. But the tasks usually aren't serial. Fixing the inventory bug, tuning the enemy AI, and writing the save system are three independent pieces of work, and there is no reason they should take turns.
What stops most people from running three at once is the verifying. Three streams of output mean three checkouts, three editors, and a lot of alt-tabbing.
Satura Space gives each agent a real terminal, a genuine shell rather than a chat box, and lets you tile them across panes, tabs and windows. The Unity session they are all changing is in the same window, streaming, so verifying a result is a glance rather than a context switch.
PATH are detected automatically; you pick per terminal at launch.Claude Code, Codex and Cursor get the Unity MCP server registered in the project automatically. An agent that only edits files cannot check its own work; a pre-wired one opens a scene, presses Play, reads the console, and takes a screenshot of what it built. The other three run as ordinary shell agents.
Two agents editing one checkout will step on each other. Through the Unity MCP server an agent can branch the project for itself: unity_worktree_create makes an isolated git worktree under .sspace/agent-worktrees/ on branch agent/<name> and boots a headless Unity editor in it. Every other tool works against that worktree once you pass its path, and finished work merges back with git merge agent/<name>.
A full worktree brings your uncommitted changes with it and seeds the Library copy-on-write, so it boots with everything present. A settings worktree is sparser: ProjectSettings/ and Packages/ with an empty Assets/, which is the sandbox to hand an agent that is building something from scratch at your project's exact Unity version. Either way the worktree sits inside .sspace/, so it stays out of the file tree, out of search, and out of the parent checkout's git status.
Long agent runs and desktop apps are a bad combination: an update prompt or an accidental quit costs you a half-finished task. Terminals here live in a detached daemon that owns the pty out of process. Quitting or updating Satura Space does not kill the agent, and reattaching replays the scrollback from a ring buffer. Claude sessions additionally cold-restore through claude --resume, so closing and reopening a terminal picks the conversation back up.
Parallel agents are only useful if you can walk away. A finished turn flashes its tab green and plays a sound; an agent stopped waiting for permission flashes amber. Both fire from tabs you are not currently looking at. Nine sounds ship with the app, including the Half-Life HEV beeps.
Satura Space never proxies prompts, never resells tokens and never asks for an API key. Agents bill through the subscription you already have; the app charges a flat fee for the bench itself. There are no credits and no overage.
Six out of the boxClaude Code, Codex, Cursor, Grok Build, Kimi, opencode.
Auto-registeredFor Claude Code, Codex and Cursor.
Detached daemonApp quit or update doesn't kill a running agent.
Bring your ownNo key, no proxying, no token resale, no credits.
Claude Code, Codex, Cursor, Grok Build, Kimi and opencode, whichever are on your PATH. The first three get the Unity MCP server registered automatically; every terminal is a real shell, so anything else works too.
No. Bring the CLI agents you already pay for; Satura Space never proxies prompts, resells tokens or asks for a key.
Terminals live in a detached daemon, so the agent keeps going. Reattaching replays the scrollback, and Claude sessions cold-restore with claude --resume.
A finished turn flashes the tab green and plays a sound; a permission stop flashes amber. Both work from tabs you aren't looking at.
For Claude Code, yes. Both are set per terminal on the launch screen and passed through as flags.
Yes, two ways. Launch each session in its own git worktree from the start screen, or let the agent make its own with unity_worktree_create, which branches the project into .sspace/agent-worktrees/ and boots a headless editor there. On macOS the Library clones instantly with no extra disk.
Parallel agent terminals, one Unity project, and the game running beside them.