You build the agent. We run it.

Give an AI agent any job, from answering your customers to writing the Monday report. Describe it in plain words, or write it as a file. Reactor runs it, shows what every run costs, and keeps it inside the limits you set, whatever a customer types.

  • Your agents only use the tools you give them
  • They stay inside the spending caps you set
  • You can see every conversation and what it cost

Example loops, running live

Each one is a real loop, and you don’t need an account to try it.

Answers questions about Reactor from its own docs, and names the page it used.

One loop, one job.

You describe the job in your own words, and we run it around the clock on a budget you set. It can sit in a chat on your site, or run on its own with no chat at all.

We run the loop. You own the data.

A few jobs a loop can do

It answers your customers’ questions

Point it at your website and documents, and it answers at 2 a.m. with a link to the right page. When it doesn’t know, it hands the chat to your team.

It spots an unhappy customer

When someone is stuck or upset, it flags the chat so a person on your team can take over.

It takes action, with a human in the loop

It can book the appointment or start the refund. Anything you mark for approval waits until someone on your team says yes.

It works on its own, too

A loop can run on a schedule, or kick off when a new order comes in. No chat needed.

An AI agent is easy to build and hard to run. We run it.

What goes wrong when you run one yourself, and what we do instead.

If you run it yourself

With Reactor

It does something you never meant it to.

It can only touch the tools and the data you hand it.

A customer talks it into something it shouldn’t do.

No message can unlock a new tool, and anything you flag waits for a yes from your team.

One heavy user burns through your month’s budget.

Every user has a spending cap, and so does every loop.

A small change quietly breaks it.

A new version only goes live after it passes the tests you wrote for it.

Nobody can say what it did last Tuesday.

You can see everything it did, and what it cost.

You host it, and you restart it when it crashes.

We host it and keep it running.

Plug in your own tools and APIs, or use the ones we’ve built in.

Three steps to your first loop

  1. Describe it

    Tell LoopAssistant, our built-in helper, what you want the loop to do.

    The console’s chat with LoopAssistant: the person types what the loop is for, to qualify the leads that arrive on a pricing page and file each one for sales.
  2. Check it

    See what it will do and what it can cost, then test it yourself before it goes live.

    A published loop on the console: what runs now, what a conversation is told, and what every run may do.
  3. Go live

    Add a chat to your site, or connect your own app through our API.

    A loop’s On your site tab on the console: the accent colour, the light or dark setting, and the code to paste into a site.

Or build it where you already work.

A loop is a plain text file, so you can build one from your terminal or from your AI coding tool. Either way, it gets the same permissions you have on the account, and no more.

From your terminal

rxr is our command-line tool for macOS and Linux. Describe the loop, then edit the file and test it in a chat as you go. While rxr is running, a loop can also run commands on your machine, but only the ones you’ve listed.

$ curl -fsSL https://plutonium.io/install.sh | sh
$ rxr login
rxr is connected to https://mcp.plutonium.io/mcp. Try: rxr loops
$ rxr new --describe "Answer questions about our return policy"
Wrote returns_helper.loop.yml: the draft of returns_helper.
$ rxr dev returns_helper.loop.yml
Saved returns_helper version 1 as the draft.
you> Can I return shoes after 40 days?
The command line guide

From the AI tool you use

It works with Claude Code, Cursor, Claude Desktop, or anything else that speaks MCP. Sign in once, and your AI tool asks you before it publishes.

$ claude mcp add --transport http plutonium https://mcp.plutonium.io/mcp
$ claude
> Build a loop that answers questions about our return policy.
I drafted returns_helper and ran the draft. It replied:
"Returns are open for 30 days from delivery."
Shall I publish it?
Connect an AI tool

You own the data

Every lead, booking and ticket is saved to your account. You can fetch them through the API, or have each one sent to your webhook or your inbox.

Pay for what your loops use

There are no per-seat fees. And when a loop keeps doing the same steps, it can offer to turn them into a plain automation with no AI model involved. Approve it once, and each run after that costs a fraction of what it did.

Reactor loop

About 0.5¢ for a typical run on the Pro plan

On pay as you go, it’s about 0.8¢. A run is one conversation, or one job a loop does on its own. You pay for what the AI model uses, plus a small fee per model call.

Reactor search

About 7¢ per 1,000 visitor searches on the Pro plan

Searches sent from your own servers are billed at a separate flat rate.

Pro is $10 a month, and it gets you the lower prices. See the full pricing.

Get on the waitlist

We’re taking on a few teams at a time. Leave your email, and we’ll let you know when your account is ready.

You’re on the list

We add people by hand. When your account is ready, we’ll email you, and you sign in with Google at that address.

One moment…

How we use these details: Privacy policy.