Integrations, the Prompt Editor & Regex Libraries

This guide gathers three power-user areas that the rest of the app leans on. Integrations is where you connect outside services. The Prompt Editor is where you shape how the AI behaves. Pattern Profiles are reusable libraries of regular expressions. You will not need them every day, but knowing what they do — and the options each offers — unlocks the rest of bAIbel AV.

Integrations — connecting outside services

Many features reach beyond your machine: translation-management servers, rented compute, local model servers, and voice input. Integrations is the single place where those connections live. When another screen tells you a service is “not set”, this is where you go to fix it.

Section	Purpose	What you set
TMS Server	Connect to a translation-management server to download assigned project packages.	The server address, your username and password, then the tasks to download.
External Servers Running	See which external compute servers are active right now.	Their type, state, uptime or cost, and a control to shut each down.
External Server Configurations	Save the settings for external compute, such as rented GPUs.	Named configurations you can create, import, and reuse.

Where your keys live

API keys are held securely by the application, not shown on screen. Local model and voice servers send no authentication, so keep them on a trusted network and never expose them to the open internet. The app reminds you of this where it matters.

Figure 1. The Integrations screen. Connect a translation-management server, watch running servers, and manage saved configurations.

The Prompt Editor — shaping the AI

Every AI task in bAIbel AV — translating, summarising, extracting terms — is driven by a prompt: the instructions the model is given. The Prompt Editor lets you read and adjust those prompts. It is a powerful tool, and a careful one: changing a prompt changes how the AI behaves across the app.

Open Prompt Editor from the sidebar. Before you can edit, give the prompt a Name and choose a Category — the category tells the editor which task the prompt belongs to. You can also set a Status of Active, Draft, or Archived, and use Load Default to start from the built-in template for that category.

A prompt has two parts, each in its own editor:

Part	What it is
System Prompt	The standing instructions that set the model’s role and rules.
User Prompt	The request itself, into which the document’s content is placed at run time.

Prompts use named placeholders, written like {SOURCE_TEXT}, that are filled in when the task runs. A Valid Terms panel lists the placeholders allowed for the chosen category, so you can insert the right ones rather than guess. The toolbar gives you the usual New, Open, Save, Save As…, Export, and Delete.

Edit with care

The built-in prompts are tuned to work well. Copy one with Save As… and adjust the copy, rather than overwriting the original, so you can always fall back to a version that works. Keep the placeholders intact — removing one can break the task that relies on it.

Figure 2. The Prompt Editor. Set a name and category, edit the System and User prompts, and insert allowed placeholders from the Valid Terms panel.

Regex libraries — Pattern Profiles

Some content is best matched by a rule rather than by AI: dates, reference numbers, email addresses, product codes. Pattern Profiles are reusable libraries of such rules, written as regular expressions, that bAIbel AV can apply — for example to recognise and protect sensitive patterns during anonymisation. You manage them under Privacy and Confidentiality, on the Pattern Profiles tab.

The feature has three layers, from building blocks to finished libraries:

Layer	What it is
Pattern Groups	The building blocks — a single regular expression for one kind of thing, such as “dates” or “email addresses”, with options and validation.
Pattern Profiles	A curated set of groups, scoped to a language, that you apply together.
Pattern Translations	Rules for carrying a matched pattern across languages — for example reordering the parts of a date for the target language.

When you build a Pattern Group, you write the expression, paste in some Sample Text, and use Analyze to check it matches what you expect before you save it. Options such as ignoring case or matching across lines are there when you need them.

When to reach for a pattern

Use a Pattern Profile when something has a reliable shape a rule can catch. It is the precise, repeatable counterpart to the AI-driven privacy tools: the Privacy Profile and Numerical Obfuscation wizard handle the judgement calls, while patterns handle the predictable shapes.

Figure 3. The Pattern Profiles tab. Pattern Groups are the building blocks, Profiles gather them, and Translations carry matches across languages.

Figure 4. Building a Pattern Group. Write the expression, test it against sample text with Analyze, and set its options before saving.

Terminology used in this guide

Integrations: The screen where you connect outside services: translation-management servers, external compute, and local model or voice servers.
TMS server: A translation-management server that assigns and exchanges project packages.
Prompt: The instructions given to a model for a task, made of a System Prompt and a User Prompt.
Placeholder: A named slot in a prompt, such as {SOURCE_TEXT}, filled with content when the task runs.
Pattern Profile: A reusable library of regular-expression rules, built from Pattern Groups, applied for example during anonymisation.
Pattern Group: A single regular expression for one kind of content, the building block of a profile.