sMind

Appearance

Download

Get started

smind ​

sMind by Spacing Mind is a self-hosted platform for running coding agents, delivered as one Go binary with a web app, desktop app, command line and mobile access. This page is for developers who run coding agents every day and want them on their own machine under their own accounts.

sMind by Spacing Mind, version 0.17.0. This page describes the 0.17.0 release.

What smind is ​

smind is a self-hosted platform for running coding agents. It is one Go binary that serves both the API and a web interface, and it spreads agent traffic across your own provider accounts instead of a single shared pool. Each task runs in its own isolated git worktree, so several agents can work on one repository without stepping on each other. You use it from the web app, the desktop app, the command line or a phone.

It is for developers who run coding agents every day and want them on their own machine or server, under their own accounts, with a clear view of what each agent is doing and what it costs.

  • Getting started: install smind, start the daemon, create your first workspace and task.
  • Providers: connect Claude, Codex, GLM and Kimi accounts.
  • Provider recipes: add a login for Z.AI, DeepSeek, OpenRouter or a local model server from a built-in template.
  • Why a provider or model is greyed out: the Signed out, Plan used up, Cooling down, Not in plan and Unsupported labels, and which ones block a send.
  • Write-only secrets: how provider keys are stored, what you can see afterwards and what is not protected.
  • Workspaces and tasks: the daily workflow, chats, attachments and the queue.
  • Git and pull requests: branches, publishing, updating from base and opening pull requests.
  • Mobile and relay pairing: pair a phone, what end-to-end encryption protects, push notifications.
  • Orchestration: run a Lead that launches Peers, with budgets, alerts and permission rules.
  • The validator seat and review rows: a fresh-context check of a commit against a contract, and its verdicts in the inbox.
  • The Steward review seat: a read-only second opinion on a finished task, with objective gates, a verdict and limits.
  • Inbox: one list of permissions, verdicts, reviews, failed runs and stalled work that need you, on the web and on a phone.
  • Scripting and permissions: drive agents from scripts, permission modes, the MCP server.
  • Permission policy: answer routine permission requests from a file, with a fixed floor of always-denied commands.
  • Run the daemon: run it as a service, reach it from other devices, Docker, updates, backups.
  • Usage and limits: what your accounts have used and what happens at a rate limit.
  • Z.AI usage meter: the 5-hour and weekly windows of a Z.AI coding plan.
  • CLI reference: every smind command.
  • Troubleshooting: fix a daemon that will not start, login problems and reconnects.
  • Configuration reference: the keys in config.yaml and their defaults.
  • Plugins: install MCP servers, skills, profiles, commands, themes and plugin processes, and see what they do first.
  • Writing a plugin: the manifest reference, capabilities, hooks, RPCs and a worked example.
  • Automations: start a task on a schedule or when a workflow fails.
  • Connected hosts: pair two daemons and choose which direction works.

How the docs are organised ​

The pages are listed in reading order: first orientation and install, then daily work, then mobile and orchestration, then reference pages you look things up in. Each page stands on its own and links to the pages it relies on. Commands are shown in copy-pasteable blocks; run smind --help for the usage text of the version you have installed.