Space for your agents,
free for you.
Manage, orchestrate and automate your AI agents, from anywhere.
curl -fsSL https://spacingmind.com/install.sh | shDrives the agents you already use:
and models from Z.AI GLM, Qwen, DeepSeek, Kimi, MiniMax, OpenRouter, Ollama, LM Studio, a LiteLLM proxy or any Anthropic- or OpenAI-compatible endpoint.
via Agent Client Protocol & native protocols
Why sMind
Running many agents shouldn’t mean managing chaos
Without:Every agent shares one credential and one rate limit.
With sMind:Per-workspace account routing.
Without:Parallel agents trample the same working tree.
With sMind:A git worktree per task.
Without:A terminal to babysit per agent.
With sMind:Detaching never stops a run. Approve from the web UI or CLI.
Without:A failed run is a wall of provider text.
With sMind:A class and a next step: sign in, wait for the reset, pick another model.
Features
Manage, orchestrate, automate. From anywhere.
Manage
Every agent in one place
Group work into workspaces and spaces. Every task gets its own git worktree and branch, so several agents can work on one repo at once, and you follow them all from one chat, diff and terminal view. One inbox lists what needs you: permissions, failed runs, reviews and stalled work.
Read the docsOrchestrate
Agents that run agents
smind mcp serve exposes sMind as MCP tools. A lead agent can create child tasks, wait on their runs, read their logs and stop them, while you keep the final say on permissions. When a task is done, ask the read-only Steward seat for a second opinion: it checks the gates, returns a verdict and never merges.
Automate
Unattended when you want it
Pick a permission mode per run, from plan-only to auto-accept. Anything the mode still escalates waits for you inline, or is answered from a policy file with a fixed floor of denied commands. Automations start a task on a schedule or when CI fails. Script everything from the CLI.
Read the docsRun command: rm -rf ./tmp/fixtures && go generate ./...
Bash
Claude Code wants to run a command
Matched rule ask-destructive
Anywhere
Your agents, wherever you are
Connect to a local daemon, any daemon you can reach by URL (say over your VPN or an SSH tunnel), or one behind a relay you host yourself with smind relay. Relay traffic is end-to-end encrypted, so it only ever sees ciphertext.
Your machines · preview
More than one box, one set of agents
Pair sMind on another machine with a six-digit code you type yourself, tag each host with labels like os:linux or gpu:cuda, and start a task on whichever one fits. Each direction is switched on separately, and every call between hosts is written to an audit log you read with smind host audit.
studioConnected
This host → studio
studio → this host
Labels
winboxConnected
Labels
Your accounts
Routed per workspace
sMind proxies Anthropic and OpenAI-compatible traffic across the provider accounts you add. Pin a workspace to one account, or share a pool with 24h session affinity and a failover chain. The model picker says before you send if a row is signed out, out of plan or cooling down, and a failed run names its class and the next step.
Read the docsClaude Code
Claude Code
One daemon, every surface
Use it from wherever you work
One Go binary runs the daemon, API and web UI. Every client talks to it over the same WebSocket API.
CLI
Linux · macOS
Manage workspaces, spaces and tasks, send prompts, attach and tail logs.
Web UI
Built into the binary
Chat timeline, diff viewer, CodeMirror editor, real PTY terminal and inline approvals at localhost:4648.
Desktop app
macOS · Windows
Native app with the daemon bundled (macOS) or installed and managed natively (Windows): auto-update, notifications, open-in-editor, and connections to local, remote or relayed daemons.
Mobile
Coming soon
Follow runs and approve permissions from your phone, over the end-to-end encrypted relay.
$ curl -fsSL https://spacingmind.com/install.sh | sh
$ smind serve # daemon + web UI on http://localhost:4648
$ smind workspace create ~/code/smind "smind" hard
$ smind task new <workspaceId> "fix the failing test"
$ smind task send <taskId> claude-native "fix the failing test"
Open source
Built in the open
sMind is self-hosted end to end: the daemon, web UI, desktop app and relay. It is early and pre-1.0, so expect rough edges and breaking changes. Issues and pull requests are welcome.
Dogfood-first
Built by people who run it every day.
No feature without real demand
Scope follows actual use, not a checklist.
Every decision stays reversible
Decisions are written down as ADRs and can be undone.
Latest release
v0.18.0
Nov 2, 2026
- smind-desktop-0.18.0-macos-arm64.dmg
- smind-desktop-0.18.0-macos-x86_64.dmg
- smind_0.18.0_darwin_amd64.tar.gz
- smind_0.18.0_darwin_arm64.tar.gz
- smind_0.18.0_linux_amd64.tar.gz
- smind_0.18.0_linux_arm64.tar.gz
- smind_0.18.0_windows_amd64.zip
- smind_0.18.0_windows_arm64.zip
- sMind_0.18.0_x64-setup.exe
- sMind_0.18.0_x64_en-US.msi
FAQ
Questions, answered
Is sMind free?
Yes. sMind is open source and runs on your own machine. There is no sMind account or hosted service to sign up for.
Does my code leave my machine?
Your repos and worktrees stay on the machine running the daemon. Agents talk to the model providers you configure, using your own accounts. The relay is optional, self-hostable and end-to-end encrypted.
Can it use more than one machine?
Yes, as a preview. Pair sMind on another machine with a code you type, label each host (for example os:linux or gpu:cuda), and start tasks on the host that fits. Each direction of a pairing is off until you turn it on, and calls between hosts are audited.
Which agents does it support?
Agent CLIs: Claude Code, Codex, GLM, Kimi, OpenCode, GitHub Copilot, Cline, Goose and Cursor, all behind one interface. Models and plans: Z.AI GLM, Qwen, DeepSeek, Kimi, MiniMax, OpenRouter, Ollama, LM Studio, a LiteLLM proxy or any Anthropic- or OpenAI-compatible endpoint.
Do I need the desktop app?
No. On Linux, macOS and Windows a single binary gives you the daemon, CLI and web UI. The desktop app (macOS and Windows) adds a native window, auto-update and a managed daemon.
Why does Windows SmartScreen warn about the installer?
The Windows installer is not code-signed yet, so SmartScreen may warn on first run. The desktop app installs, updates and restarts the daemon for you.
Is it ready for daily use?
The daemon, CLI and web UI are used day to day, but sMind is pre-1.0: expect rough edges and breaking changes before v1.
Spacing Mind services
Rolling out agents across a team?
The people who build sMind work as forward-deployed engineers: we set it up with your team, wire it into your workflow, and ship alongside you.
Give your agents some space.
curl -fsSL https://spacingmind.com/install.sh | sh