Run it from your phone
Chapter 5 of Your own AI agent, from anywhere. New: tested on our own machines, so tell us how it goes.
Everything is in place. This chapter is about using it from where you actually are: on your phone, away from your desk. There's nothing new to install.
What you need
- Chapters 1 to 4 done.
- Your phone, with the Claude app signed in to the same Claude account.
1. Talk to your agent
On your phone, open your agent's address in the browser and sign in with Edgible, just as on your computer. Then add it to your home screen (Share, then Add to Home Screen on an iPhone; the menu, then Add to home screen on Android), so it opens like an app.
Ask it anything, as you would on your computer:
Find three good walking trails near me for Saturday, and tell me which suits a beginner.
It's the same agent, with the same memory and the same budget, wherever you open it.
2. Check on everything from Claude
In the Claude app, start a new chat. The Edgible connector from chapter 2 is already there. Ask:
Is everything on my Edgible working?
Claude checks each app answers the way it should, and tells you if anything needs attention.
3. See what your AI really costs
Show my AI spending, by agent and by model.
You get the real dollars: each agent, what it spent on which model, and how much of its budget is left. This is the owner's view, which only you have. Each agent sees only its own.
4. Change things, safely
From your phone, Claude can also change things. It asks you first every time:
Raise my agent's budget to ten dollars a month.
Claude shows you the old and new budget, and what your agent has spent so far, before it changes anything.
If a change needs a key, Claude sends you a key link instead of asking you to paste it into the chat.
What you've just done
You've done the heart of the playbook. You have:
- Your own AI agent, on your own machine, that you can talk to from anywhere, your phone included.
- Guardrails that work without you watching: every agent has its own key and its own budget, and it stops at its limit.
- The real dollars, any time you ask, by agent and by model.
- Everything behind your own sign-in. No ports open, no keys in chats, and every change asked first and recorded.
- Nothing locked in. It's all open source, on your machine, and each step can be done by hand.
Most people meet AI agents through someone else's service, on someone else's terms. Yours is yours.
One finishing touch: A home for everything you've built.
Prefer to do it by hand?
Your agent is also in its own page, in any browser. Spending and budgets are also in the gateway's admin page, under its usage view and each agent's key, for anyone who has the gateway's master key: when you set the gateway up by hand, you do. With Claude's connector it is kept for you and never shown, so spending and budgets are one question to Claude away, as in this chapter.