Book a call

AI agents & automation studio

AI that handles the busywork, quietly and correctly.

Loomwork designs and builds AI agents that answer customers, move data between your tools and close the loop, with your team approving anything that matters.

  • Built on the tools you already use
  • Every action logged and explainable
  • People stay in charge
Support agent · live preview Online
Hi! My order arrived damaged. Can I get a replacement?
⌕Found order #20418 · delivered 2 days ago
✓Replacement policy: eligible within 30 days
Sorry about that! I've sent a free replacement. It ships tomorrow, and you'll get tracking by email.
Replacement created · Ticket closed · Logged to CRM
Can I move my Thursday appointment to next week?
⌕Checked calendar · 3 open slots next week
✓Customer prefers mornings (from past visits)
Of course. Tuesday 10:00 or Wednesday 9:30 are free. Which suits you?
Waiting for reply · Old slot held for 2 hours
Please send me last month's invoice again.
⌕Verified sender matches account email
✓Found invoice for October in billing
Done! The October invoice is in your inbox now. Anything else I can help with?
Invoice re-sent · No human needed
Example conversations · demo only

Connects to the tools you already run on

EmailCalendarsCRMHelpdeskSpreadsheetsPaymentsTeam chatDocs & wikisPhone & SMSAccountingE-commerceForms

How it works

From a message to a finished job, in three steps.

  1. 01

    Listen

    A request arrives by email, chat, form or phone. The agent reads it and works out what the person actually needs.

  2. 02

    Decide

    It follows your rules, written in plain language, and checks your systems for the facts before it does anything.

  3. 03

    Act

    It does the work across your tools, replies, and logs every step. Anything risky waits for a person to approve.

Workflows

One agent, many kinds of work.

Pick a kind of business to see a typical workflow we'd build. Each step lights up as the agent runs it.

Running
TriggerNew order emailinbox · orders
AI stepRead & sortintent: return request
ActionCreate return labelshipping
ActionRefund to cardpayments · under limit
ResultReply + update CRMcustomer notified

Example workflows · demo only

The console

See every decision it makes.

Each agent comes with a simple console. You can see what it did, why it did it, and what's waiting for you. No black box.

OverviewThis week
Example data
Requests handled0example
Solved without a person0example
Waiting for approval0example
Avg. first reply0example
Requests by dayhandled · escalated
Mon
Tue
Wed
Thu
Fri
Sat
Sun
Activitylive
  • Done

    Refunded duplicate charge

    2m
  • Approve

    Refund above limit

    6m
  • Done

    Rescheduled appointment

    11m
  • Done

    Updated CRM contact

    14m
  • Asked

    Requested missing order number

    20m

You approve what matters

Set limits once. Anything above them waits for a person, with the full context ready.

Every step is logged

A plain-language trail of what the agent saw, decided and did. Easy to check, easy to undo.

Your data stays yours

Agents only reach the systems you connect, with the least access they need.

Under the hood

Every run, written down step by step.

Behind each finished job is a plain record: what came in, what the agent checked, which tools it used and what it decided. Your team can read it, search it and replay it.

  • ReadablePlain words, not code. Anyone on the team can follow it.
  • ReplayableRe-run any job on the same input to check a change.
  • Versioned rulesEvery rule change is saved, so you know what applied when.
run #a7f3 · invoice processingExample run
  1. 09:41:02triggerinbox.attachmentinvoice_0381.pdf from Harbor Paper Co.
  2. 09:41:02extractinvoice.readtotal 1,240.00 · 6 line items · due in 30 days
  3. 09:41:03matchpurchase_orders.findPO-2291 ✓ 6 of 6 lines match
  4. 09:41:03checkrules.apply"total within PO amount" ✓
  5. 09:41:04postaccounting.bill.createdraft bill saved
  6. 09:41:05notifyteam_chat.send#finance · "Invoice 0381 is ready to pay"
  7. 09:41:05logaudit.writerun saved · can be replayed
✓ Done in 3.1sWaiting for payment approval

Guardrails

You decide where the line is.

Set a limit once. Below it, the agent just gets on with it. Above it, the job waits for a person, with everything they need to say yes or no. Try it.

$100
  • Refund a delivery fee$12Runs on its ownWaits for you
  • Re-send a lost parcel$38Runs on its ownWaits for you
  • Refund a damaged item$64Runs on its ownWaits for you
  • Goodwill credit$120Runs on its ownWaits for you
  • Refund a full order$240Runs on its ownWaits for you
  • Cancel a yearly plan$480Runs on its ownWaits for you

3 of 6 run on their own · 3 wait for a person

Example actions and amounts

Integrations

Plugged into everything you already run on.

Requests come in from wherever your customers are. The agent does the work inside the systems your team already uses. We build, test and watch every connection, so nothing gets copied by hand.

Email
Website chat
Forms
Phone & SMS
CRM
Payments
Calendar
Helpdesk

What we build

Four kinds of agents. One careful way of building them.

Where's my order?On its way. Arriving Friday.Thank you!

Customer support agents

Answer questions, track orders, handle refunds and returns, all in your brand's voice, any hour of the day.

FormCheckCRMInvoice

Back-office automation

Move data between forms, spreadsheets, CRM and accounting without copy-paste or missed steps.

Team knowledge assistant

Your team asks a question in plain words and gets an answer from your own docs, with the source linked.

Phone & booking receptionist

Picks up calls, books appointments into your calendar and sends a text confirmation.

Process

Small first, then steady improvements.

We start with one job that wastes your team's time, prove it works, then grow from there. Timings below are a typical example.

1Week 1

Map

We sit with your team, list the repetitive jobs and pick the one with the clearest payoff.

2Weeks 2–4

Build

We write the rules with you, connect your tools and build the agent and its console.

3Week 5

Test together

It runs alongside your team on real requests. People approve every action until you trust it.

4Ongoing

Launch & tune

It goes live with limits you set. We review the log with you each month and keep improving it.

FAQ

Good questions to ask.

What if the agent gets something wrong?

You set limits on what it can do alone. Anything above them goes to a person first, and every action is logged so it can be checked and reversed.

Do we need to change our software?

No. We connect to the tools you already use. If a tool has no way to connect, we'll tell you early and suggest options.

How long until it's useful?

Usually a few weeks for the first agent, because we start with one clear job instead of everything at once.

Is our customer data safe?

Agents get only the access they need. We agree in writing where data goes and how long anything is kept.

Will it replace our team?

It takes the repetitive part of the job so your people can spend their time on the cases that need a human.

Start small

Tell us one job you'd hand to an AI tomorrow.

We'll reply with an honest view: whether it's a good fit, roughly how it would work, and what it would take.

Demo form: nothing is sent.