rply

ntfy for agents. One inbox between you and your AI agents. Free to start at app.rply.sh, open source and self-hostable.

Website · Docs · Hosted app · @rply/cli on npm

Any agent with a shell (Claude Code, Codex, a cron script, CI) installs the rply CLI and can then:

• send you updates (markdown, tables, progress bars, images, files)
• ask questions and wait for your answer (choices, forms, confirmations, free text)
• pin live-updating status messages and full-page HTML apps to the top of its channel
• request secrets that go straight into a file on its machine, never through its context or the server
• get woken up when you write to it

You read and reply in one inbox for all your agents: a web app served by the server, and a mobile app (Expo, in progress).

rply is agent-agnostic and not a session mirror: it doesn't control or host your agent, it gives the agent a way to talk to you.

How it works

agent machine                    your server                 you
rply CLI  ── agent token ──▶  rply server (Docker,  ◀── your account ──  web / mobile app
                              SQLite, serves web app)

• One channel per agent (or per project). Agents hold agent tokens (CLI); you sign in to your account on every device and see all your channels there.
• Agents join and channels get claimed with one-time codes, so no long-lived credential ever ends up in the agent's output.

Quick start

1. Sign up at app.rply.sh and create a channel.
2. Paste the setup prompt the app shows into your agent. It installs the CLI, joins the channel and sets itself up.

That's it; your agent can now message you. Prefer to set it up by hand? See Connect an agent by hand.

Self-hosting

The server is one container with an SQLite database. Keep /data on a volume.

git clone https://github.com/joshuaKnauber/rply && cd rply
docker build -t rply .
docker run -d -p 8080:8080 -v rply-data:/data -e RPLY_URL=https://rply.example.com rply

Put it behind HTTPS (Caddy, Railway, exe.dev's proxy, …). Run one replica only. On Railway, .railway/railway.ts defines the service, healthcheck and /data volume; generate a domain and set RPLY_URL to it in the dashboard.

Open the server URL in a browser and create your account: the first person to register owns the server. After that, registration follows RPLY_REGISTRATION. Without email configured, reset a forgotten password on the server with docker exec <container> node packages/server/src/reset-password.ts <email>, which prints a new one.

Connect an agent by hand

On the agent's machine (Node 20+):

npm install --global @rply/cli
rply join https://app.rply.sh <code>   # or your own server's URL

Create the channel in the app (Add channel): it shows a setup prompt with a one-time agent code to paste to the agent.

Or let the agent start: rply init https://app.rply.sh --name "SEO bot" creates a channel and prints a one-time claim link (and a QR code in a terminal). Open it while signed in to add the channel to your account. The link expires after an hour (rply pair prints a new one), and a channel nobody claims is deleted after 24 hours.

Teach the agent

rply guide setup     # installs a background service that runs app scripts and wakes the agent when you write
rply guide           # commands, blocks, HTML, apps, secrets, files
rply guide --snippet >> CLAUDE.md   # or AGENTS.md

For agents reading this

If you are an AI agent and your human wants to talk to you through rply:

1. Install the CLI: npm install --global @rply/cli (or use npx @rply/cli).
2. Connect: rply join <server-url> <code> if they gave you an agent code, otherwise rply init <server-url> --name "<your name>" and give them the claim link it prints. Without a server of their own, use https://app.rply.sh.
3. Run rply guide setup and follow it, then read rply guide.
4. Add the output of rply guide --snippet to your CLAUDE.md / AGENTS.md.

The basics:

rply send "Deployed **3** pages"
rply ask "Publish these drafts?" --choices Publish,Skip --wait --timeout 2h   # prints the answer
rply inbox                                    # new messages from the human
rply ack <id> working|done
rply status set status.json                   # pinned status, a JSON blocks array
rply secret request OPENAI_API_KEY --to .env --reason "for keyword research" --wait
rply send "Weekly report" --image chart.png --file data.csv

Never put secrets in messages; use rply secret request. Add --json to any command for machine-readable output. Exit code 2 means timeout, 3 means cancelled. Full CLI reference: packages/cli/README.md.

Development

Node 24+ (the server runs TypeScript directly, no build step).

npm ci
npm run dev:server     # API on :8080, rply.db in the cwd
npm run dev:web        # web app on :5173, proxies /v1 to :8080
npm run dev:mobile     # Expo app
npm run rply -- <cmd>  # CLI from source
npm test && npm run typecheck

CLAUDE.md has the full design notes (architecture, API, data model, decisions).

License

The server, web app, mobile app and website are licensed under AGPL-3.0. The CLI (packages/cli) and the shared schema package (packages/schema) are MIT, so you can use and embed them anywhere.