AI Pool Build Prompt
Build Your Own Local AI Account Pool
One prompt for your coding agent. It checks what's already on your Mac, installs the proxy, sets up account rotation, and builds the system from there. It doesn't guess, and it changes nothing until you approve it.
No email required. Copy it, paste it into your coding agent, and the first pass is read-only discovery.
- Read-only discovery first
- Approval before changes
- Automatic rotation, retry, and cooldown
- Localhost-only architecture
- Built-in testing, repair, and uninstall steps
01The deliverable
What This Prompt Builds
A local AI proxy that only listens on your own machine and holds as many provider accounts as you add. It spreads work across them, retries a failed request on a different account, and cools down an account that hit its limit. The core build has no interface. It runs headless as a background service, and adding another account later takes one login command.
The core
The proxy engine is the core deliverable. Everything else is optional.
If your agent builds only the pool, you already have the point of the whole thing: rotation, retry, and cooldown across every account you authorize. It runs quietly in the background with nothing else installed on top of it.
Rotation you configure, not code
Every credential you authorize becomes a candidate. The proxy picks among them, hands a failed request to another one, and skips an account that's cooling down. The prompt is blunt about this: don't write rotation logic yourself. The proxy already does it.
Loopback only, secrets kept quiet
The service binds to 127.0.0.1 and nothing else. The remote control panel stays off and the config file is owner-only. No token, key, or account identifier gets into chat, logs, screenshots, or the finished build.
Optional layers
Everything Below Here Is Optional
The prompt keeps these separate on purpose. Stop once the pool works and you still have the part that matters. Add any of these only if you actually want them.
Optional
Quota-aware routing guard
A plugin that replaces the stock pick order with a deliberate drain order. It doesn't add rotation. Rotation already works without it.
Optional
Headless usage observer
A small background process that reads how much capacity each account has left and publishes a non-secret snapshot for the guard. No window, no UI.
Optional
Terminal launchers
Namespaced commands that point your terminal sessions at the local proxy. The official commands stay untouched.
Optional
T3 Code configuration
Optional editor setup so your coding tasks run through the same proxy, models, and limits as your terminal.
Optional
Status app
A read-only view of what the pool is doing. It displays. It never picks an account. Skip it and you lose nothing.
Optional
Doctor, repair, uninstall, tests
A read-only health check, narrow repair actions, a clean removal path that runs twice safely, and tests for the rotation cases that matter.
It makes the agent tell you the truth about both modes
The prompt forces a choice before anything is built. Supported-only mode uses supported API keys and documented profiles, and still tells you to check your providers’ current terms before routing any consumer login through a proxy. Full-fidelity mode pools several paid consumer subscriptions, which is how you keep working past the limit any one of them sets, and which likely breaks individual-use and account-sharing terms. You are the one who has to read your own providers’ current terms and pick.
02Setup
How To Use This
- 1
Click Copy Full Prompt
Take the whole thing. The safety rules are load-bearing, so don't trim it down.
- 2
Open a new task in your coding agent
Claude Code, Codex, or anything else that can read your Mac and run commands.
- 3
Paste the complete prompt
One task, one paste. The prompt tells the agent how to behave before it tells it what to build.
- 4
Let it perform read-only discovery
It looks at what's already installed, which ports are taken, and what would clash. It changes nothing.
- 5
Review the proposed architecture and approve the build mode
You get a plain-English plan and a straight choice. Nothing gets installed until you answer.
03The whole thing
The Full Prompt
Nothing is held back. The complete text is below, exactly as your agent will read it. It's long on purpose. Most of that length is about how the agent should behave, not what to build.
04What we actually do
This Is The Kind Of Infrastructure We Build To Run Outbound At Scale
We built this prompt because the systems behind serious outbound are more technical than most people realize.
Our founder rotates 7+ Fable 5 accounts himself, the same kind of pool this prompt builds.
Our processes handle large lead volumes and send more than ten million emails every month for clients. That takes infrastructure, automation, deliverability engineering, data systems, and campaign operations that all work together.
The AI pool prompt is not our service. It's one small example of the technical work we do behind the scenes every day.
Our actual service is done-for-you cold email: we help B2B companies map their market, build the infrastructure, test offers and creative, and turn outbound into a measurable acquisition channel.
Explore the outbound system behind the campaigns.