Deploy Paperclip (w/ CloudCLI)

Scheduled AI agent jobs with run history, plus Claude Code in any browser.

Deploy Paperclip (w/ CloudCLI)

Paperclip

Just deployed

/paperclip

/var/lib/postgresql/data

Just deployed

/home/node

Deploy and Host Paperclip (w/ CloudCLI) on Railway

User manual: setup, jobs in git, updates and troubleshooting, with screenshots.

Claude on a schedule, for any recurring task, with every run on record. Connect any GitHub repo you can push to, even one unrelated to the jobs or made just for them, and tell Claude in plain words what should happen and when. Claude writes the job's instructions into that repo, schedules the job in Paperclip and tests it once. From then on Paperclip runs it on time and keeps every run. Claude Code in your browser comes with it, on your own Claude subscription.

Paperclip (w/ CloudCLI) banner

About Hosting Paperclip (w/ CloudCLI)

Paperclip with a private Postgres, and CloudCLI with Claude Code, Codex, git and the GitHub CLI, behind one login. Your own Claude account (Pro or Max) signs in through Claude Code's official sign-in, and the token stays on your server.

Why it's different

Most AI agents keep what you set up in their own memory, on their own server. Here, what a job does and when it runs are files in your GitHub repo. Paperclip runs them; Claude does the work.

  • Yours to keep. A job's instructions are a skill file, its schedule an entry in .claude/jobs.json, both in your repo. Stop using this template and your jobs stay: plain text another scheduler or agent can run, which one command recreates on a new server.
  • Traceable. Every change to a job, what it does or when, is a commit you can review or revert.
  • Auditable. Every run is on record in Paperclip: status, result, full transcript, model and tokens.
  • Predictable. Only scheduled jobs run, one task per run: the Runner does what the skill says, closes the task and stops, or marks it blocked with the reason instead of guessing. In the chat, Claude asks before anything beyond file edits and basic git and GitHub commands.
  • Easy to start. An email and a password to deploy, a page to connect Claude and GitHub, then as many repos and jobs as you need.
  • Open. Open-source apps on your server, and this template's code in a public repo.

Not only for code

Git keeps the jobs; it doesn't limit what they do. A job can be anything Claude Code can do from a server: summarize a web page, check that your site answers, build a report, draft a weekly update. A job that needs a key gets it from the Runner's environment variables in Paperclip.

Quickstart: from deploy to your first job

From a fresh deploy, with a demo repo: a coffee shop's website with three open issues.

1. Deploy

1. Press Deploy Now.

Step 1: Deploy Now

2. On Paperclip, press Configure and type your email and a password (8 to 128 characters): your login for both apps, and the whole form. Press Save Config.

Step 2: email, password, Save Config

3. Press Deploy. The first start takes a few minutes.

Step 3: Deploy

4. Once all three are online, open the CloudCLI service's URL.

Step 4: the CloudCLI service

2. Connect your accounts

CloudCLI opens its setup page until you finish it; redo any step later at /setup.

5. Sign in with the email and password from the deploy form.

Step 5: setup page sign-in

6. Claude: press Start: the page runs Claude Code's own claude setup-token and shows its link. Open it, sign in, approve, then paste the code and press Submit.

Step 6: Claude sign-in link and code

7. Connected: CloudCLI and Paperclip's Runner both use it, with no redeploy.

Step 7: Claude connected

8. GitHub: press Start, enter the code at github.com/login/device and authorize. Both apps can then use your repos, private ones included.

Step 8: GitHub code

9. Your repo: choose the one for your jobs and press Use this repo.

Step 9: choose the repo

10. The repo is cloned for both apps, and the time zone comes from your browser. Press Finish and open CloudCLI at the bottom.

Step 10: repo and time zone set

3. Ask Claude for a job

11. CloudCLI opens a chat in your repo. New Session starts another.

Step 11: a new chat in CloudCLI

12. Not sure where to start? Say hi: Claude explains how jobs work and suggests some for your repo.

Step 12: Claude suggests jobs

13. Ask for a job in plain words, with when it should run. Claude writes its instructions as a skill in your repo (here .claude/skills/open-issues-digest/SKILL.md), schedules the job in Paperclip and records it in .claude/jobs.json.

Step 13: asking for a job

14. Claude runs it once as a test and reports the result, with links to Paperclip.

Step 14: the test run result

4. Follow and change your jobs

Open the Paperclip service's URL, with the same login.

15. Routines: each job's schedule, next and last runs, and an on/off switch.

Step 15: the job in Paperclip

16. Each run is a task, where the Runner posts its result.

Step 16: a run posts its result

17. Agents, Runner, Runs: every run's status, model, tokens, duration and full transcript.

Step 17: a run with its tokens

18. To change a job, ask Claude: a new time or instructions, pause, resume, run now, or show the last log.

Step 18: changing a job by chat

Good to know

  • Jobs are files in your repo. The skill says what to do, .claude/jobs.json when; Paperclip runs them. Edit a skill with Claude or by hand and push: the Runner pulls before every run.
  • A new server: connect the same repos on /setup, then ask Claude to restore your jobs.
  • Cost. Each run is a Claude Code session on your subscription. The Runner uses Sonnet: the digest above took about 86,000 tokens, mostly cached. Change the model on the Runner's page.
  • More repos. Add each on step 3 of /setup; a job belongs to the repo of the chat you create it in.
  • Shared memory. A .claude/memory/ folder in your repo becomes Claude's memory in both apps.
  • A blocked job usually lacks a login; its task says which. Redo that step at /setup.
  • New password: change ADMIN_PASSWORD on Paperclip, then redeploy Paperclip and CloudCLI.

Common Use Cases

  • Daily or weekly reports from web pages, APIs or data files
  • Checks on a schedule: a site that should answer, prices or listings to watch, links that break
  • Repo upkeep: issue triage, dependency checks, release notes
  • Claude Code sessions from any browser or your phone, on your own server

Dependencies for Paperclip (w/ CloudCLI) Hosting

  • Paperclip
  • CloudCLI
  • Postgres

Deployment Dependencies

Implementation Details

Paperclip (w/ CloudCLI) architecture

Needs Hobby or above (about 1.4 GB of memory at idle). Paperclip dumps its database hourly to its volume; also turn on Railway backups for the three volumes, and keep PAPERCLIP_SECRETS_MASTER_KEY, which encrypts Paperclip's secrets.

Why Deploy Paperclip (w/ CloudCLI) on Railway?

Railway is a singular platform to deploy your infrastructure stack. Railway will host your infrastructure so you don't have to deal with configuration, while allowing you to vertically and horizontally scale it.

By deploying Paperclip (w/ CloudCLI) on Railway, you are one step closer to supporting a complete full-stack application with minimal burden. Host your servers, databases, AI agents, and more on Railway.

Updated 2026-11-02: jobs recorded in each repo's .claude/jobs.json, restorable on a new server; several repos; a user manual. Paperclip 2026.1023.0 and CloudCLI 1.37.7.


Template Content

More templates in this category

View Template
Chat Chat
Chat Chat, your own unified chat and search to AI platform.

okisdev
117
View Template
Weaviate
Vector database with hybrid search and an API key-protected API

GitHub Locker
1
View Template
Möbius Personal AI OS
Self-hosted AI agent that builds and runs apps for you, with memory.

Tracklayer
1