Skip to content

Agentic operating system

AI agents that move your company’s processes forward, on your own server

When a request comes in through your website, an invoice lands in your inbox or a scheduled task comes due, an agent handles it on a server reserved for your business. Anything that commits you is set up to wait for confirmation from a person on your team, the operator.

Request a diagnosticSee how it works

An example day

  1. 7:52 a.m. Supplier invoice received by email, checked against the purchase order

  2. 9:40 a.m. Quote request on your website, summary confirmed by the end customer

  3. 1:15 p.m. Order received by webhook, confirmation to the end customer drafted

  4. 4:30 p.m. Bank statement dropped in a folder, reconciliation ready for your accountant

Who does what

The four actors in a Belowdecks instance

The Horizon symbol works as a legend. The operator decides above the line, the agents work below it, your customers write in from outside and the Belowdecks team stays behind the scenes.

  1. Step 1 of 4: Marc writes in and confirms his request. Who acts: Marc, End customer
  2. Step 2 of 4: The agents prepare the quote. Who acts: Agents
  3. Step 3 of 4: Julie approves sending. Who acts: Julie, sales, Operator
  4. Step 4 of 4: The agents deliver the quote. Who acts: Agents
Julie, Marc and the company Entretien Boréal are fictional. The quote goes out through a sending function or a mail connection set up with you at installation.

  • Marc, End customer

    Outside the mark,Writes in and confirms their own request

    The end customer is a visitor to your website, a customer, a supplier or a job candidate. In a conversation, they can confirm the summary of their own request, and that ends their part.

    • I need a quote for three buildings.
    • Yes, that’s my request.

  • Julie, sales, Operator

    Above the line,Decides

    The operator is the person in your company who runs Belowdecks. They receive the agents’ questions, approve their work and decide what goes out. Several people on your team can share this role.

    • Confirms sending
    • Answers from an email

  • Agents

    Below the line,Works

    Autonomous AI agents work on the instance reserved for your company, each named for its role. They prepare and draft, ask a question when the decision belongs to the operator, then finish with a verdict.

    • Front-desk agent
    • Quoting agent

  • Belowdecks team

    Behind the scenes,Sets up and monitors

    We install your instance, set up the agents, functions and connections with you, then monitor how it runs. We never make decisions on your company’s behalf.

    • Installs the instance
    • Sets up the agents with you
    • Monitors

Processes

Your company’s processes, from prospecting to after-sales

We group them into five business cycles, from first contact to financial statements. Each cycle has its own page: what starts the work, what the agents do and where the operator confirms.

Always on, self-evolving, proactive

What the agents do on their own, and when they bring in the operator

Agents run on your dedicated instance without waiting for someone to write to them. Here is how they start, how they get better from week to week and how they reach the operator, the person in your company who runs Belowdecks.

  • Always on

    Work starts when the event happens, nights and weekends included

    An email from an end customer, an article in an RSS feed, a file dropped in a folder, a schedule written in plain words, a signed webhook from one of your systems or new rows in a collection put an agent to work. Nobody needs to open a conversation, and the instance that does the work serves your company alone.

    • When an engine account hits its limit, the run can carry on with the next one
    • A run cut short by a crash is detected, marked as an error and can resume
    • A master switch lets the operator pause every run at once
    • A backup every night, kept on the same server

  • Self-evolving

    Every week, the instance proposes fixes and spots repeated work that could move to code

    Every night, an analysis with no model cost spots repeated failures, timeouts, unusual costs and forgotten approvals. A time limit that is too short is extended right away when the diagnosis is certain; other insights arrive on Monday in the operator’s conversation, with a card when the fix is precise. After a successful mission, an agent can save its mechanical steps as functions, which then run with no token cost.

    • Project memory and search across past runs
    • AI analysis of a problem’s root cause, agent by agent
    • An agent can propose a change to its own mission, and the operator decides
    • Functions an agent saves after its mission are tried on examples first

  • Proactive

    When information is missing, the agent asks the operator and picks up again with the answer

    The operator answers in the instance or from the email that brings the question, without signing in, and your systems can answer through the API. Notification rules alert the operator and your team by email or webhook to a failure, a pending approval, a question or an insight, and an agent can alert someone during its run.

    • The weekly review, every Monday in the operator’s conversation
    • A usage report by email on the 1st of each month
    • A cost forecast from past runs before you switch on a scheduled agent
    • Approve, answer or confirm from a link in an email

Demo

Three examples at Entretien Boréal, to move forward one step at a time

Each step shows who acts and where: the end customer on your website or by email, the agent on your instance, the operator in their inbox or in the instance. Each step’s button moves the example on to the next one.

Who acts in this example

  • End customer Outside the mark: writes in and confirms their own request
  • Operator Above the line: decides
  • Agents Below the line: works
  • Belowdecks team Behind the scenes: sets up and monitors

Marc asks for a quote on the website, and Julie approves it from her email

Lead-to-Cash cycle at Entretien Boréal. Marc, the end customer, confirms the summary of his request, and that ends his part. The quoting agent prepares the work, and Julie, the operator, decides what goes out.

Belowdecks · A request on your website

Example

The company’s website

Marc, Immeubles Rive-Sud, End customer End customer Now
Marc, Immeubles Rive-Sud, End customer

Marc writes here, on the company’s website.

Message ready to send: Hello, I’d like a quote for the maintenance of our three office buildings in Longueuil.

Inbox

Julie, sales, Operator Operator
Julie, sales, Operator

Julie receives the agents’ questions and cards here.

Below the line

Agents Agents
Agents

The agents work here, below the line: they prepare, compare and draft.

Step 1 of 8 Marc, Immeubles Rive-Sud, End customer The company’s website

In these examples, emails to end customers and suppliers go out through a sending function or a mail connection set up with you at installation. The people and companies are fictional.

A chat assistant and Belowdecks, point by point

Both rely on AI models. The table compares categories of tools and targets no particular product.

  • What starts the work

    A chat assistant
    Your message
    Belowdecks
    An email, a schedule, a webhook, a data change

  • Availability

    A chat assistant
    For the length of the conversation
    Belowdecks
    Around the clock, on your dedicated server

  • What it does

    A chat assistant
    It answers and drafts
    Belowdecks
    Agents carry out the steps of a process

  • Memory

    A chat assistant
    The current conversation, sometimes your preferences
    Belowdecks
    Project memory and past runs

  • Binding actions

    A chat assistant
    You copy the answer and act yourself
    Belowdecks
    Set up to wait for the operator, even by email

  • Missing information

    A chat assistant
    It asks you during the conversation
    Belowdecks
    The agent asks the operator and waits

  • Improvement

    A chat assistant
    As the vendor ships updates
    Belowdecks
    Insights every night, fixes put to the operator on Monday

  • Costs

    A chat assistant
    Usually a subscription per person
    Belowdecks
    Subscription plus tokens, with an optional cap per run

  • Your data

    A chat assistant
    Stored with the service’s vendor
    Belowdecks
    Stored on your dedicated instance

Presentation

The Belowdecks presentation video

A video of under 40 seconds, with a voice-over and captions, hosted on this site. The company and the people it shows are fictional.

Text of the video

On a Friday at 9:47 p.m., Nadia sends a quote request to three companies. Two will reply on Monday. At the third, a Belowdecks agent replies within a minute with a link to book a call, and Nadia books a meeting for Monday morning. The reply goes out through a sending function set up at installation.

The video then shows what the agents did over the same weekend: 9 quotes prepared, 4 customers followed up and 2 orders confirmed. The quotes wait for approval from a person in the company, who approves them on Monday.

It ends on the five cycles that Belowdecks moves forward, from sales to accounting, and on the invitation to request a diagnostic.

Connections

How the instance connects to your company

Your mailbox, your website and your software stay as they are. The instance connects to them through standard means, configured at setup.

  • Email

    Each new message in a watched mailbox starts an agent. Notices, reports and agents’ questions also go out by email, to the operator and your team.

  • Your website assistant

    One script tag adds it to your pages. It answers end customers in their own language and can have them confirm their own request.

  • Files

    A file added or changed in a watched folder becomes a task.

  • Schedules

    Every Monday at 7 a.m. or on the 1st of the month, set on screen in plain words.

  • Webhooks

    Your software calls a trigger’s address, and its signature is checked. The instance in turn sends signed webhooks to your tools.

  • API and MCP

    Your systems start agents and follow their runs through the REST API or the MCP server. Agents reach your software through its API or an MCP server.

  • Collections

    Structured data that agents fill in and read back: requests, orders, invoices. Your software can write to them through the API.

  • Team space

    The operator and your team work with the agents in simple screens, within the rights of their group.

Control and security

What your company controls on its instance

Four points that hold for every instance. The Security page covers each one in detail.

  • An instance dedicated to your company

    A server, a database and accounts that serve you alone. No data passes from one client company to another.

  • Your Anthropic account and your spending limit

    Tokens are billed by Anthropic directly to your account, with a limit you set. Each agent can also be given a cost cap per run.

  • Agent confinement, switched on and verified at setup

    Each run sees only its own working folder on the server. The Belowdecks team switches this confinement on and checks it when each instance is set up.

  • Export at any time

    You can export your projects (agents, skills, functions, collections) and, if you wish, the rows of your collections, up to a cap per collection. An export never contains the values of your secrets.

First step

Let’s choose the first process to hand to your agents

The diagnostic looks at your actual work and finds the cycle to start with. You then receive a written plan.

Request a diagnosticSee the offer