Agents, apps, and your data
live in one space.

A self-hosted home for your AI, on one server you own

> Install ai-space on my server following github.com/ericz-lab/ai-space/blob/main/docs/install-by-agent.md

Paste it into Claude Code or Codex. The agent installs, stops at the logins, and tells you what to click.

Web panel iPhone macOS Linux space CLI
✦ Search apps, or tell Space Assistant what to do
Inbox
Shortcuts
Todo
Notes
Portfolio
Todo Planner
Terminal
Space Assistant
AI Usage · Today
1.24M tokens across 3 machines
Todo · Up next
  • Review the backup report
  • Approve the rebalance draft
  • Read the morning digest
Space Ops
Backups verified on all 3 peers
  • Services running
  • Next task

One system instead of a dozen tools

You use it through three things

Apps

Hold structured information and the scenarios you work in, and give the AI lasting context.

Agents

Take on complex, unstructured requests and carry out the work, as Claude Code or Codex sessions with their skills and memory.

Todo · Up nextReview the backup report09:30

Widgets

Keep the important state and results visible at a glance, on the panel, the iPhone and the Mac.

What nobody wants to build twice

Every app gets a core it can lean on

Scheduler

at / every / cron schedules and event triggers, with http, command and agent targets.

Bus

Events and calls between apps, delivered at least once, with a catalogue and a history.

Notify

One call reaches Telegram, Feishu, Slack, Discord, Bark and more; queued, retried, gathered in an inbox.

Model calls

One endpoint for every app, a concurrency cap, and a token ledger by app, purpose and model.

Storage & backup

Per-app SQLite or PostgreSQL and a blob store; daily snapshots to S3, weekly verification.

Supervision

Each app's service runs as a systemd unit or LaunchAgent, kept in step with its manifest.

Peers

One panel over several machines: a GPU box, a server abroad, the Mac on your desk.

Terminal & CLI

A web terminal behind your login, and a space command for everything the API does.

From idea to running app

Ask your coding agent. ai-space does the rest.

  1. 1

    Install by agent

    A written procedure your agent follows: one directory, plain files and SQLite, your own domain behind a login.

  2. 2

    Describe an app

    The shared space-app skill scaffolds it to the app spec: a space.yaml, a service, agents, widgets, tasks.

  3. 3

    It lives on

    git push deploys it. It runs, schedules, reacts to other apps and pings you when something matters.

space.yaml
name: morning-brief
tasks:
  - name: brief
    schedule: "0 8 * * *"
    run: { agent: { prompt: prompts/brief.md } }
    notify: { when: [ok], channel: telegram }

Give your AI a home that keeps running.

Open source, self-hosted, early stage. Bun and TypeScript, one server, one directory.