Skip to content
Client portal
EN
EnglishΕλληνικάRomână

AI that helps your team, with a person in charge.

Assistants that answer from your own documents, and automation for repetitive work. Replies to customers go out only after a person approves them, and we agree how to measure success before launch.

Start this projectTalk it through first

Damaged on arrivalAwaiting your approvalSent at 14:07Reply due in 15 min

Cracked frame on arrival

Elin Hedin[email protected]4 orders since 2021

Elin Hedin

My Arvik 3 frame arrived yesterday with a crack next to the bottom bracket. I have a race on the 4th.

Can you replace it quickly, or should I ask for a refund?

IMG_4471.jpg

Draft reply

Written by AIEdited by Mira

Hi Elin,

I am sorry the frame reached you cracked. You can choose a free replacement or a full refund, and we pay shipping both ways.1

MB

We can send a new frame within 5 to 7 working days by express today, so it reaches you on Thursday, well before your race.2

Race on the 4th, so I switched to express.

MB

Good luck at the race!

Mira, Norrvik Cycles

We can send a new frame by express today, so it reaches you on Thursday, well before your race.2

Race on the 4th, so I switched to express.

MB

Good luck at the race!

Mira, Norrvik Cycles

Sending as Mira BergEscalate⇧EApprove and send⌘↵Sent to Elin HedinUndo

Confidence0.85

  • Replace or refund
  • Express by Thursday
  • Shipping paid

Sources

1

Returns policy§ 4.2

Damaged items are replaced or refunded, shipping paid both ways.

2

Live stockAR3-54-SG

3 in stock. Express before 15:00 arrives in 2 days.

Crack, 11 mm

Arvik 3 gravel bike€1,249

54 cm, Bronze · NC-20931

Delivered 21 Sep

Checks

  • Matches returns policy 4.2
  • Order details match
  • Send before 15:00 for express

Automations

412 conversations today

35.2%edited by a person

0sent without a person

Classify412
Retrieve374
Draft361
Human review361
Send344223 as drafted121 edited
Aurora DeskEvery reply is approved by a person

Aurora DeskA support desk where AI drafts and a person approves. Fictional business, built to show the work.

What to look at

What this covers

Every service in this family.

Systems that interpret information or carry out defined workflows, using models together with conventional software. 20 services

  • Website and support assistantsAnswers from approved information, with a hand-over to a person.
  • Internal knowledge searchSearch company documents with source references and access controls.
  • Document extraction and OCRExtract structured fields from supported invoices, forms and files.
  • Document comparison and reviewCompare versions, summarize material and flag items for review.
  • Transcription and conversation analysisSpeech to text, summaries by speaker and action items.
  • Translation and localization workflowsPrepare translations and route sensitive content for review.
  • Lead qualification and routingClassify enquiries and send them to the right workflow.
  • Sales and email assistanceDraft responses and coordinate authorized follow-up.
  • Report and content generationProduce structured drafts from approved sources and data.
  • Workflow and administrative automationAutomate repetitive tasks with APIs, rules and approval steps.
  • Agentic multi-tool workflowsCoordinate bounded tasks across connected tools and services.
  • API, MCP and plugin integrationsExpose controlled tools to supported AI assistants and clients.
  • Voice and multimodal interfacesSpeech, text, images and supported real-time interactions.
  • Image, video and media pipelinesIntegrate licensed generation and processing tools into workflows.
  • Classification and anomaly workflowsLabel material or detect unusual activity for review.
  • Recommendations and forecastingRanking and prediction from data, where suitable data exists.
  • Local and private model deploymentOn-device or hosted inference, where the hardware and model suit it.
  • AI evaluation and optimizationMeasure quality, speed, cost, retrieval and failure behaviour.
  • Model adaptation and fine-tuningAdapt supported models when the evidence justifies the investment.
  • Human review and agent supervisionReview queues, approvals, traces and intervention controls.
Technology options

Python and TypeScript are common choices for orchestration; Rust, Go or C++ can support specific processing needs. The work uses model APIs or suitable local runtimes, retrieval stores, queues and evaluation tools. Where plain rules are enough, no AI is used.

How we choose technology

Questions

Which AI models do you use?

The model is chosen for the task, the cost and your privacy needs. It can be changed later without rebuilding the system.

Will our data be used to train models?

We choose providers and settings that keep your data out of model training, and we name every service that handles it.

What if the AI gets something wrong?

It is designed to say when it is unsure and hand over to a person. Where mistakes matter, a person approves before anything is sent.

Is this the same as AI search and SEO?

No. That service is about how AI assistants recommend your business. This one builds AI into your own products and daily work.

Start

Tell us what you are building.

A short description is enough. You get a straight read on scope, cost, and our honest take on the best approach, before anyone commits.

Start a projectSee the packages

Write to [email protected] · New projects and general questions

  • Built from zero, no templates
  • Hosted and maintained by the same team
  • Scope and cost agreed before we start
  • Updates, monitoring and support after launch
  • Findable in search and AI answers