Always-on assistant infrastructure

Put your coding agent on call.¶

Push turns Claude Code, Codex, or Pi into a personal assistant you can message and schedule. It runs on your machine, keeps durable state, and sends results back to iMessage or Telegram.

Get started See the architecture

Choose a path¶

Run Push for the first time

Install the binary, connect one channel, configure a backend, and validate the setup.

Quickstart
Connect your chat

Set up private iMessage or Telegram conversations with narrow sender allowlists.

Configure channels
Automate recurring work

Write Markdown runbooks, run them manually, or add cron triggers and send stored results to your primary chat.

Jobs and schedules
Operate it continuously

Choose permissions, inspect local state, and run Push under launchd or systemd.

Operations guide

The mental model¶

Push is a gateway, not an agent runtime:

message or cron trigger
        ↓
Push: filter → route → persist → schedule → deliver
        ↓
Claude Code, Codex, or Pi: reason → use tools → produce result

You own one Git-versioned assistant repository containing identity, context, and jobs. Push owns channels, history, scheduling, approvals, security, and delivery. The selected backend owns models, tools, MCP servers, skills, and authentication. That boundary keeps Push small and lets the backend change without rebuilding your assistant.

Documentation map¶

If you need to…	Read…
install and run one working channel	Quickstart
understand every TOML setting	Configuration
add recurring or manual work	Jobs and schedules
choose backend permissions safely	Permissions and security
keep Push online after logout or reboot	Run as a service
inspect commands and outputs	CLI reference
understand or extend the code	Architecture and contributing

Canonical source

These pages are generated directly from the Markdown in the repository's docs/ directory. If the site and source ever disagree, update the Markdown source and rebuild the site.