Max
$24a month
Parallel agents and heavy builds
- vCPU
- 8
- RAM
- 16 GB
- disk
- 200 GB
- Internet
- 200 Mbit
Start the seven-hour job. Close the laptop
Backfills, scrapers, embedding runs and batch scripts keep going in a tab on your box, next to the database they feed. Max gives them 8 vCPU and 16 GB for $24 a month.
From $5 a month · Docker and sudo included · Cancel from the dashboard any time
Your laptop was never built for a seven-hour run
Close the lid and the job dies at batch 212 of 640. Leave it open and the machine is hot, slow and tethered to a charger all day.
A cloud notebook times out. A bare VPS costs an evening of setup before the first row moves.
One long job
- 09:10
Start it in a tab
Run the script in the browser terminal or over SSH. The tab is a tmux session on the box.
- 09:11
Walk away
Close the tab, the laptop, the train Wi-Fi. The job keeps writing to the database in Docker next to it.
- 13:00
Glance at it from your phone
Open the dashboard on your phone and look at the tab: batch 320 of 640, on schedule.
- 16:01
Pick up the results
Query them on the box, or forward the database port to your laptop and open it in your own client.
What a box gives a long job
Hours of runtime, no charger
The job runs on the box's CPU and memory. Your laptop stays free for everything else.
Data next to the code
Postgres, Redis or Jupyter run in Docker on the same box, so the job never crosses the internet for its own data.
Disk for real datasets
Up to 200 GB on Max. Keep raw dumps, parquet files and intermediate tables on the box.
An agent on night watch
Hand the script to Codex or Claude Code: it runs it, reads the traceback when a batch fails and restarts from the last good one.
The box for this job
Max · $24
8 vCPU and 16 GB hold a heavy pandas job, its database and an agent in memory together. Lighter jobs fit on Pro.
- Box
- etl
- Plan
- Max · 8 vCPU · 16 GB · 200 GB · $24/mo
- Agent
- Codex
- Tabs
- 2
- Services
- postgres:18 · jupyter
- Runs
- Around the clock, lid closed
Questions about this
Is there a GPU?
Boxes are CPU machines with up to 8 vCPU and 16 GB. They fit ETL, scraping, backfills, API-based embedding runs and evaluation jobs. Train large models on a GPU cloud and let the box run everything around it.
What if the node reboots mid-job?
Planned reboots come with 48 hours' notice by email, so you can time a long job around them. Files and Docker volumes survive; running processes start fresh, so give long jobs a resume point.
How do I get the results out?
Forward a port over SSH, serve them at a preview URL, or push them to your own storage straight from the box.
The parts of the box this job uses
More jobs for a box
Start the seven-hour job. Close the laptop
Start a box, hand the agent the job and close the tab. Come back to finished work.
Start my boxCharged when the box starts · Cancel from the dashboard; the box runs to the end of the paid month