PlainAI

PlainAI · the AI arm of TheFaceCraft

AI agents and automation, engineered in Switzerland.

We build AI agents that take work off your team, inside the CRM, inbox, ERP and finance tools you already run, under rules your business agrees before anything goes live. Based in Lucerne, working across Europe and Asia.

What we build

What we build, and who for

Three kinds of agent, running inside the systems you already pay for. We work with owners and operations teams in companies of every size, from a ten-person firm to a group of several thousand, across Europe and Asia.

Price follows the complexity of the system, not the size of your company. A smaller build has fewer integrations, fewer rules and fewer failure paths, so it costs less to build and less to run.

Customer

Support and triage

Reads, classifies and answers customer messages across email, shared inbox, WhatsApp Business and live chat. Order status, bookings, returns and delivery questions. Anything outside its rules goes to a named person with the full thread attached.

Revenue

Lead qualification and follow-up

Scores inbound enquiries against your own criteria, writes the follow-up, books the call and keeps CRM records clean. Every conversation is logged against the right record, so nothing depends on a rep remembering to type it up.

Operations

Invoice processing and reporting

Document extraction from invoices, delivery notes and contracts, reconciled against your ERP or accounting system. Stock and scheduling updates, plus the recurring report written from your own data instead of a spreadsheet.

Difference

Why our agents behave differently

A generic agent answers from the model’s defaults. Ours answers from a page of rules your business writes and signs off before a line of code exists.

Point at a rule to see the words it produced. Change the rule and every reply the agent sends changes with it.

The rules page
What the agent sent
I can’t price this before we scope it. Thursday 09:00 is open, and Marc in operations will run the call.

Process

How a build runs, and how long it takes

Four steps. You can stop after any of them and keep what you have.

The same four steps whichever agent you start with. See the services.

60 min

Workshop

One hour with the people who do the work: how decisions get made, where the time goes, which cases must never be automated. You leave with the rules page.

1 week

Specification

What the agent reads, what it writes, how it decides, and where it stops for a human. It names the model and the cloud: Claude, GPT or Gemini, on Bedrock, Azure or Google Cloud. Price is fixed against it.

Weeks

Build

Running on your real data, in your accounts, from the first week. Orchestrated with n8n, LangGraph or agent code, traced in LangSmith or Langfuse, with spend caps and human approval on anything uncertain.

Day one

Handover

Repository, prompts, workflow definitions, credentials and documentation, plus a training session for whoever runs it. Our access is removed. Support afterwards is available and optional.

Company & data

Where we are, and where your data sits

The company behind PlainAI, and the accounts and jurisdiction your agents run in.

Company

Registered in Switzerland

FaceCraft GmbH, Grossmatte-Ost 24b, 6014 Luzern, entered in the commercial register of the Canton of Lucerne.

Data

Your accounts, your jurisdiction

Agents run in your own cloud tenancy on AWS Bedrock, Microsoft Azure or Google Cloud, so data residency stays where your policy requires. Nothing is used to train public models. Work is done to the revised Swiss Federal Act on Data Protection (nFADP) and the GDPR, and the specification lists every system the agent can reach, with an audit trail of what it did.

We build with

AnthropicOpenAIGoogle GeminiClaude CodeLangChainCrewAIHugging Facen8nModel Context ProtocolGoogle CloudNVIDIAUiPathBraintrustPostgreSQLGitHubPythonWhatsApp Business

Start

Start with the workshop

One hour on your business, before anything is built. You leave with the rules page and a shortlist of the work worth automating, whether or not we build it.

FAQ

Questions people ask

The ones that come up before a first call.

PlainAI is the AI arm of TheFaceCraft, a Swiss branding strategy and perception firm registered as FaceCraft GmbH in Lucerne, Switzerland. We design and build AI agents that run inside a company’s own systems: reading and writing in its CRM, shared inbox, ERP and finance tools, deciding by rules the business agrees in advance, and escalating to a named person where the rules do not cover the case. Every build is handed over with its repository, prompts, workflow definitions, credentials and documentation, and runs in the client’s own cloud accounts.

Price follows the complexity of the system, not the size of your company. A build with two integrations, a short rules page and one escalation path costs less than one spanning six systems with approval steps and audit requirements, because it is less work to specify, build, test and run. The price is fixed against the specification rather than billed by the hour, so it is known before the build starts. The workshop comes first because until the work is mapped, any number is a guess.

No. We work with owners and operations teams in companies of every size, from ten-person firms to groups of several thousand, across Europe and Asia. The build is scoped to the work, so a small company gets a smaller system with fewer integrations and a lower price rather than a cut-down version of something built for someone else.

The systems you already run. Agents reach your tools over the Model Context Protocol (MCP) and over standard APIs, covering CRMs, shared inboxes, WhatsApp Business, calendars, ERP and accounting systems, document stores and databases such as PostgreSQL. Where a workflow runner fits better than code, we use n8n, whose catalogue covers several hundred services natively. Connecting a new system is usually a scoping question, not a rebuild.

In your own accounts and your own cloud tenancy, on AWS Bedrock, Microsoft Azure or Google Cloud, so data residency stays wherever your policy requires. We build around your data instead of moving it. Nothing is used to train public models. Work is done to the revised Swiss Federal Act on Data Protection (nFADP) and to the GDPR, and the specification lists every system the agent can reach before it reaches one, with an audit trail of every action it takes.

Four things, all decided before launch. The rules page defines what it may and may not do. Human-in-the-loop approval sits on any action above a threshold you set, such as issuing a refund or sending a quote. Spend caps limit model and API cost. Every step is traced in LangSmith or Langfuse, so a wrong answer can be read back and the rule corrected rather than guessed at.

The people who build the system are the people in the workshop. Whoever writes your specification writes your agent, so nothing is lost between the conversation and the build, and there is no account layer between you and the work.

The repository, every prompt, the workflow definitions, the documentation and the credentials, all in your own accounts. The system runs whether or not we stay involved. Ongoing support exists because some teams want it, not because the system needs us to keep running.