# AI agents you can put in front of customers.

Your customers ask. Your loop answers, acts, and tells you what matters. You decide what it can read, do and spend. Reactor enforces those limits on every run, whatever a visitor types.

[Read the docs](/docs.md)

- Only the tools you grant
- Spending caps per loop, per user and per run
- Every run and every cent on your console

## [Reactor loop](#loop)

[Agents that answer your customers, act in your tools, and tell you what matters.](#loop)

[How a loop works](#loop)

## [Reactor search](#search)

[Search that understands the question, for your visitors, your team and your loops.](#search)

[Try the search](#search)

## A loop is an AI agent that does one job, inside limits you set.

plutonium.io Reactor runs AI agents for your business, on your site and in your tools, around the clock. You say what the loop should do. We run it safely, on a budget you set, and everything it collects is yours.

We run the loop. You own the data.

### What a loop does for you

#### Answers from what you already wrote

Point it at your site and docs. It reads them, stays current when they change, and answers at 2 a.m. with a link to the right page. When it does not know, it hands the conversation to your team.

#### Catches an unhappy customer before they leave

It reads every message, files the problem, and tells your team when someone is stuck or upset.

#### Does the work, not only the talking

Books the appointment, qualifies the lead, files the expense, starts the refund. Anything that needs approval waits until your team says yes.

#### Works on its own, where you allow it

It can run on a schedule, or when an event arrives, with nobody in the chat. It sorts the inbox, summarises the week, or acts on a new order, always inside the limits you set.

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

You describe what the loop is for. We handle everything after that.

| Run it yourself, and you own these problems        | With Reactor                                                               |
| -------------------------------------------------- | -------------------------------------------------------------------------- |
| It calls a tool it should not, or the wrong one.   | It can use only the tools you grant. Nothing a user types can add another. |
| A hostile message tricks it into a harmful action. | An action that needs approval waits until your team says yes.              |
| One busy user spends your month’s budget.          | Spending caps per loop, per user and per run, checked before each call.    |
| A prompt change quietly breaks it.                 | A new version goes live only after its test conversations pass.            |
| Nobody can say what it did last Tuesday.           | Every run, step and cent is on your console as it happens.                 |
| You host it, retry it and restart it.              | We handle the hosting, the models, the retries and the restarts.           |

Connect your own tools and APIs, and use ours when you need them.

[How the limits work](/docs/guide/permissions-and-trust.md)

## How you start

1. ### Describe it

   Tell LoopAssistant what the loop is for, in plain words.

   ![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.](/landing/describe-light.webp)

2. ### Check it

   Read what it will do, what it may use and what a run may cost. Try it before anyone else does.

   ![A published loop on the console: what runs now, what a conversation is told, and what every run may do.](/landing/publish-light.webp)

3. ### Put it live

   Add a chat to your site, or connect your own app through our API. Watch every run from your console.

   ![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.](/landing/watch-light.webp)

Or build it from a terminal, or from the AI tool you use. Install the [`rxr` command line](/docs/guide/command-line.md) with `curl -fsSL https://plutonium.io/install.sh | sh`, or [connect an AI tool](/docs/guide/connect-an-ai-tool.md).

## You own the data

Every lead, booking, ticket and answer is saved as a record in your own collections. Read them over an API, have each one sent to your webhook or your inbox, or watch them arrive on your console. A loop reads only what you give it.

## Reactor search understands the question.

Text and semantic search over your docs, your pages and your own records, for your loops, your team and your visitors. Inexpensive enough to give every visitor a search box. Pick an example index and compare the ways it searches.

## Pay for what your loops use

You do not pay per seat. Work that repeats can become an automation that runs without an AI model. You approve it once, and each later run costs a fraction of the first.

### Reactor loop

About 0.5¢ a typical run on Pro

A run costs the tokens its model uses, plus a small fee per call: about $0.005073 on Pro, or $0.007805 on pay as you go.

### Reactor search

About $0.0689 per 1,000 visitor searches on Pro

A search by a visitor or a loop costs a fraction of a cent. A search through a search key has its own flat price.

[Full pricing](/pricing.md), and what Pro costs.

## Get on the waitlist

We’re opening Reactor to a few teams at a time, and we add each one by hand. Tell us where to reach you, and we’ll email you when your account is ready.