cloud & subscription
Agents that keep watch with the laptop shut.
Agents on your own machine are free forever and think on a model subscription you already pay for. A plan adds agents that run on our servers instead — same thread, same guardrails, same decision log — thinking on our model, so there is no key of yours to bring and nothing to leave switched on.
Loading plans…
Your machine, or ours
The agent does not change. Only the question of who has to be awake for it to work.
| Your machine | Cloud | |
|---|---|---|
| Runs when the computer is off | No | Yes |
| Thinks on | Your Claude, ChatGPT or GPU | Our model — nothing to connect |
| Costs you a model subscription | Yes, your own | No |
| Guardrails, stops and decision log | Identical | Identical |
| Readable from the phone | Read-only | Full control |
| Your Robinhood token | Stays on your machine | Sealed to the worker key |
| Counts against a plan limit | No — unlimited | Yes |
What “cloud” actually means
- 01It is the same agent. One thread, readable from the desktop app and your phone. Moving an agent to the cloud changes where it runs, not what it is.
- 02Your broker token never leaves sealed storage. A cloud agent trades through a grant sealed to the worker’s key. The phone never holds a Robinhood token at all.
- 03Guardrails are identical. Order caps, the daily-loss lock, enforced stops and the decision log are engine rules — they do not relax because nobody is watching.
- 04A plan is a ceiling, not a bill. Each account runs on a metered key capped for the month. It cannot overspend it, so there is no usage invoice to be surprised by.
- 05Missed wake-ups are reported, not silently run. If a scheduled run is more than a few minutes late, the agent tells you it missed rather than trading on stale intent.
- 06The limits are enforced where they cannot be edited. Agent count, wake-up cadence and daily runs are checked in the database and in the runner — not in the app, where a patched build could talk its way past them.
Questions
- What happens when I hit the daily wake-ups?
- The agent waits for tomorrow rather than spending more of the month. It tells you in the thread; nothing is lost, and a scheduled run that is skipped is reported, not silently dropped.
- What happens if I cancel?
- Cloud agents stop at the end of the period you paid for. Nothing is deleted — the threads, the book and the decision log stay exactly where they are, and moving an agent back to your own machine takes one change of provider.
- Do I need a plan at all?
- No. Agents on your own machine are free and unlimited forever, thinking on a Claude or ChatGPT subscription you already pay for, or on a model on your own GPU. A plan buys one thing: agents that keep going when the laptop is shut.
- Can a cloud agent spend more than the plan?
- No. Every wake-up passes the plan’s daily limit and the account’s monthly ceiling before it costs anything, and the key it runs on is metered and capped at the provider. There is no usage invoice — a plan is a ceiling, not a bill.
- Is a cloud agent allowed to trade differently?
- It cannot. Order caps, the daily-loss lock, allowed symbols, market hours and engine-enforced stops are checked before any order reaches the broker, in the same code, wherever the agent runs. See guardrails.
- What model does it use?
- Ours, included with the plan — you bring no key. It is the only option for a cloud agent, because the whole point is that nothing of yours needs to be reachable for it to run.
What this is not. HarborBots is software you operate. It does not provide trading signals, tips or investment advice, and it does not promise returns. Trading involves risk, including loss of principal. You are responsible for every agent you arm.