PLC ONE functional map.
PLC ONE is a harness: a set of tools built around LLMs that makes AI actually work inside a company.
PLC ONE is a harness: a set of tools built around LLMs that makes AI actually work inside a company. It brings chat, RAG, agents and multi-agent orchestration together in one platform that runs on the customer's servers. Every project is an independent ecosystem of models, tools, integrations, permissions and billing. This is vertical orchestration, tailored to a specific team, rather than one set of tools for everyone.
The AI adoption curve
Companies go through successive stages of working with AI. PLC ONE brings them all together in one platform.
Employees learn to talk to AI. Chat is the gateway to the models and the first step of adoption.
Available2RAGConversation extended with company knowledge: procedures, invoices, contracts. Answers drawn from document fragments, with the source.
Available3AgentThe model uses tools to complete a task: sandbox, files, integrations, artifacts.
Available4Agentic AIA group of agents shares the task, the context and the models. A workflow framework for multi-agent work.
In progressFour questions before deployment
Where is our data?
The application is self-hosted and data is stored on the customer's side. Models can be chosen as cloud, local or hybrid, all the way to a full air gap with local models.
Deployment and environmentWho has access to what?
Users from LDAP and Entra ID, roles configured in the UI from more than a hundred permissions, set separately in each project.
Users, identity and auditWho controls the agents' work?
Reading data is allowed; writing always waits for “Confirm”. Every change and every login goes into the event log.
Conversation with the assistantHow much does it cost?
Full billing in every project and a view one level up for the whole platform, with limits for the organisation, groups and individuals.
Analytics, costs and administrationPlatform areas
Assistants and how they are built
Every project is an independent assistant: its own system prompt, knowledge base, models, tools and users. An assistant is taken from the library or built in a conversation with the wizard.
Conversation with the assistant
Chat is the gateway to AI, so it has to be simple and trustworthy: answers with sources, visible tool work, voice and files, and writing to systems only after clicking “Confirm”.
Knowledge and documents
RAG extends the conversation with company knowledge. Our own pipelines check the relevance of fragments before they reach the model, and when the documents hold no evidence, the assistant does not guess.
Files, artifacts and Skills
An agent is the model's ability to use tools. The sandbox, file generators, artifacts and skills make the assistant hand over a finished work result, not just text.
Integrations and external sources
Integrations run in the customer's network, not in the cloud. The administrator enables them per project and the user signs in with their own account, so the AI sees only what that person can see.
AI models and compliance
The model is chosen per project. The catalogue shows ZDR and the EU region, and the requirements also apply to models working in the background: during indexing, summarisation and speech.
Users, identity and audit
Security is about access: users from LDAP and Entra ID, roles configured in the UI from more than a hundred permissions, and a log that shows every change.
Analytics, costs and administration
Every project has full billing, and the whole platform has a view one level up. You can see who used AI, when and at what cost, and limits keep the budget in check.
Agents and tasks
The next step on the adoption curve: a workflow framework for multi-agent work and tasks that run on their own, within an agreed budget.
Data protection (DLP)
A hardware and software component that censors data sent to external models. Personal data stays with the customer and the model works on pseudonyms.
Deployment and environment
The application is self-hosted. The target is a closed appliance: updates without access to external registries, data and backups in the customer's storage.
Key strengths
- An AI assistant created in conversationThe wizard drafts a seven-part prompt and the configuration, and the assistant is created after approval.
- Answers that can be verifiedCitations to the exact fragment, marking of general knowledge, a relevance gate, no quoting from memory.
- Working in the organisation's systemsMicrosoft 365, Zoho, Atlassian, ServiceDesk Plus, Morpheus, Veeam, vCenter and any system via MCP, on the user's account.
- A work result, not just an answerDocuments, spreadsheets, presentations, artifacts and Skills executed in the sandbox.
- Reliable work with dataSpreadsheet operations, tables from documents, structure of large documents, descriptive attributes, Deep Search.
- Compliance and cost controlZDR and EU per model, more than a hundred permissions, audit, AD and Entra ID, event log, limits with blocking.
- One platform, many assistantsIndependent assistants, a configuration library, white-label, PL and EN, quality parameters tuned in-house.
3.1 AI assistants · 3.2 Building assistants
Assistants and how they are built
Every project is an independent assistant: its own system prompt, knowledge base, models, tools and users. An assistant is taken from the library or built in a conversation with the wizard.
Available in version 1.7
13 functions · 2 newAI assistants
Many independent assistants
Any number of assistants in one installation, each with its own name, description, icon, prompt, knowledge base, models and user list.
System prompt
The administrator defines who the assistant is, how it answers, which sources it uses and what it does not do.
Working with or without a knowledge base
Answers from the organisation's documents, or conversational work on the model's general knowledge and tools.
Visual identity
An icon and a highlight colour, a gallery of nearly 900 symbols in three colour variants, a custom welcome message.
Assistant groups
Permissions granted in bulk with a role for selected groups, a preview of effects and conflicts before saving, analytics calculated for the group.
Group tabs on the selection screen
The user sees a tab for each of their assistant groups; after returning from a conversation, the last selected one opens.
Temporary deactivation of an assistant
Access switched off while data and billing history are kept, re-enabled at any time.
Limit on the number of assistants
A deployment-wide limit blocks creating and activating further assistants; the basis for the offer packages.
Building assistants
Library of ready-made assistants
A catalogue of configurations with a description, prompt, knowledge base settings and models; search, preview, creation in a single action.
Instant library updates
The library is updated from a central repository periodically and immediately after changes are published.
Assistant wizard
An agent holds a conversation with the administrator and drafts a seven-part prompt and the configuration: models, functions, data protection requirements, groups, integrations, Skills.
Control over the wizard's work
Answers chosen from the actual configuration, a running summary, a prompt for approval, a report of reservations, resuming an interrupted conversation.
Catalogue of supporting agents
A separate section of agents that support administration; it currently contains the Assistant wizard.
3.3 Interaction with the assistant
Conversation with the assistant
Chat is the gateway to AI, so it has to be simple and trustworthy: answers with sources, visible tool work, voice and files, and writing to systems only after clicking “Confirm”.
Available in version 1.7
24 functions · 4 newAnswer
Answer as it is being written
Content appears as it is produced, with information about the actions performed in the background.
Resilience to session interruption
The answer is produced on the system side; after a refresh, a change of device or a lost connection, the work continues.
Stopping and completing an answer
Stopping generation and completing an interrupted answer without resending the query.
Answer variants
Editing a sent question or regenerating an answer creates parallel conversation variants you can switch between.
Model selection in the conversation
A model chosen from those made available to the assistant, marked with its capabilities and compliance requirements.
Four work modes
Extended reasoning, web search, short answers and Deep Search, within the scope allowed by the administrator.
Answers chosen from a list
Assistant questions with ready-made options, multiple choice and a free-text answer; questions can be skipped.
Mathematical notation
Formulas in correct notation, with protection against treating amounts as expressions.
Clear model error messages
Overload, timeout, provider unavailable, model withdrawn, insufficient funds, moderation, authentication, each with a hint.
Verifiability and transparency
Citations to sources with preview
In-text citations, a list of sources and a preview of the exact fragment; marking of unused sources and of the model's general knowledge.
Citations to web sources
Page title and address next to the citation, browsing several sources for one statement, opening in a new tab.
No quoting from the model's memory
A document fragment is quoted only once it has been found in the knowledge base; invalid citations are removed.
Visible work progress
Information about the actions being performed and a collapsible reasoning trace with its duration.
Tool work history
The course of the answer step by step: query, source check, tool rounds with the status and result of each call.
Operation approval
Before changing data the assistant shows the scope of the action and waits for consent; mode per operation: allowed, requires approval, blocked; a record of the decision in the conversation.
Files, voice and collaboration
Attachments in the conversation
Images, documents, spreadsheets, PDF, data files, audio and video recordings, with background preparation and preview.
File preview and quoting fragments
Documents, presentations and spreadsheets previewed inside the conversation; a selected fragment of a file or answer attached as a quote.
Conversation files panel
All files uploaded and produced in the conversation in one place.
Materials under the answer
Documents, spreadsheets, presentations and images prepared by the assistant, ready to download.
Voice conversation
Dictation, reading answers aloud and a speech-only conversation mode; speech models and voice set globally and per assistant.
Sharing conversations
A frozen snapshot behind a link for assistant members or signed-in users, read-only or with continuation, with or without sources.
Working on the phone
Conversation, the assistant selection screen and artifacts adapted to small screens and touch.
Organising conversations
Automatic titles with editing, summarising very long conversations while keeping the context.
Other interface elements
Conversation panel, copying, a list of Skills, a combined prompt to connect accounts, budget warnings, light and dark mode, PL and EN, five layouts of the selection screen.
Development roadmap
1 in progressRoadmap items have no fixed date, and their scope may change.
Mobile app
Conversation, assistant selection, voice and artifacts on the phone.
3.4 Knowledge base (RAG) · 3.5 Data analysis
Knowledge and documents
RAG extends the conversation with company knowledge. Our own pipelines check the relevance of fragments before they reach the model, and when the documents hold no evidence, the assistant does not guess.
Available in version 1.7
16 functions · 1 newInput and organisation
Document input
Documents, spreadsheets, presentations, PDF, data files and scans, one by one, in bulk or as a folder with subfolders; status, stop, retry, paused processing.
Read report and page retry
A partially processed document with the failed page ranges indicated and a retry for those pages only.
Library folders
Nested folders, moving, copying and deleting multiple items, search, list or tile view.
Recycle bin with automatic cleanup
Restoring to the original location, permanent deletion, retention cleaned daily; three separate permissions.
Copying content between assistants
A document or folder copied to one or several assistants in the background, under a separate global permission.
OCR of scanned documents
Content recognition that preserves the document structure.
Descriptive document attributes
Attribute values, e.g. counterparty or contract number, read from the content and available as a filter; bulk completion after the list changes.
Secure file access
Links with limited validity; files from a conversation available only to its owner.
Search and answer
Conversation narrowed to folders
For a single conversation you indicate the library folders the assistant uses.
Multi-stage search
Intent and concepts, question variants, hypothetical answer, hybrid search, re-scoring of fragments, relevance gate.
Deep Search
Breaking the topic down into sub-questions, assessing completeness and follow-up questions up to a threshold; parameters global and per assistant.
Structure of extensive documents
Split by headings, a multi-level table of contents, reading only the part needed, extracted tables.
Independent search by the assistant
A list of documents, full-text search, opening a document, filtering by attributes, reading the surrounding context.
Communication style
No disclosure of technical information and no narrating failed attempts.
Structured data
Spreadsheet analysis
Separate indexing of spreadsheets, a data description and real operations: sums, averages, groupings, filters; a separate model for describing spreadsheets.
Analysis of tables from documents
Tables extracted from documents analysed like spreadsheet data.
Development roadmap
2 in progress · 1 planned for developmentRoadmap items have no fixed date, and their scope may change.
Large file sets
Connecting shares and archives without copying: current documents with OCR and an index, the archive without vectorisation, cost at query time. Module card 05.
Requires: Local OCR on GPU, Local embedding
Changes in documents
Editing documentation by conversation, version history, a file of the new version, questions about the change history. Module card 06.
Very large PDFs
Streaming and paged processing in the background with progress, batch indexing, memory limits, first pages available while processing continues.
3.6 Document generation · 3.7 Skills
Files, artifacts and Skills
An agent is the model's ability to use tools. The sandbox, file generators, artifacts and skills make the assistant hand over a finished work result, not just text.
Available in version 1.7
12 functions · 3 newDocuments and artifacts
Ready-made files
Word-processor documents, PDF, multi-sheet spreadsheets, presentations, text files and images; formats and limits set per assistant.
File generation in the sandbox
With the sandbox enabled, files are produced programmatically, with any layout, charts and illustrations.
Images in the conversation and in files
Generating images for the conversation and illustrations for documents, enabled independently.
Artifacts
Pages, diagrams, charts and components next to the conversation: versions, preview and code, enlarging, pointing at an element, downloading.
Sharing an artifact with the team
Private by default; shared with the assistant's users via a link and revoked at any time.
List of artifacts
An overview of the assistant's artifacts with the author, versions, visibility and a jump to the conversation; a warning about an external address.
Skills and sandbox
Skills
The organisation's procedures with a description for the assistant and the user, a shortcut invocation and materials; run by the user or the assistant.
Bulk import of Skills
Individual descriptions or packages with materials, with a report for each item.
Skill export
A Skill with its materials downloaded as an archive or a single file.
Executable Skills in the sandbox
A computational procedure in an isolated environment produces an output file; the sandbox is enabled globally with a connection test, and per assistant.
Sandbox security
Isolation without system privileges, a list of allowed addresses, limits on time, volume, number of files and concurrency.
Assigning Skills to assistants
A shared set, enabled individually for selected assistants.
Development roadmap
12 planned for developmentRoadmap items have no fixed date, and their scope may change.
Session sandbox
A persistent environment for the whole conversation with a durable file system, suspending and restoring state, CPU, RAM and disk limits; tools for working on files.
Sandbox base image
Office and PDF tools, OCR with Polish language support, fonts with Polish characters; no network by default, protection against malicious files.
File manifest in the prompt
The model knows the type, the number of pages, sheets or slides and a summary of the file, and reads the content on demand.
Requires: Session sandbox
Formats not supported by the model
Presentations, archives, binary files and legacy .doc, .xls, .ppt formats: conversion, extraction, ZIP archives with Polish file names.
Requires: Session sandbox, Sandbox base image
Editing Office files
Edits to DOCX, XLSX and PPTX in their own format, tracked changes, formula recalculation and error checks before the file is released.
Requires: Session sandbox, Sandbox base image
PDF conversions and operations
Office to PDF, DOCX and Markdown both ways, PDF to images, HTML to PDF; merging, splitting, rotating, forms, watermark.
Requires: Session sandbox, Sandbox base image
Drag-and-drop files
Several files at once onto the input field and the conversation window, with a preview before sending.
Extended artifact interface
Scope to be determined.
Import Skills from ZIP
Structure validation and registration in the assistant's catalogue after human approval.
Skills created by AI
A Skill's structure generated from a task description or materials, tested in the sandbox, saved with versioning after approval.
Requires: Session sandbox, Import Skills from ZIP
Explicit invocation of Skills
Asking for consent on a match, selection next to the chat field, forcing it in the conversation; mode per assistant: ask, automatic, manual only.
Skills catalogue interface
Sorting and filtering the catalogue; form to be determined.
3.8 Integrations · 3.9 Access to external information
Integrations and external sources
Integrations run in the customer's network, not in the cloud. The administrator enables them per project and the user signs in with their own account, so the AI sees only what that person can see.
Available in version 1.7
15 functions · 1 newThe organisation's systems
Common integration principle
The administrator configures the connection and the permission ceiling, the user connects their own account; the assistant sees only what that person is entitled to.
Microsoft 365
Mail, contacts, calendar and team messaging; a three-step configuration wizard with a connection test.
Zoho
CRM with enforced bulk queries, mail and calendar; permission scope per service, all regions.
Jira, Confluence and Compass
About 35 operations via the Atlassian MCP server; the user's account or a shared credential for on-premises installations.
ServiceDesk Plus
More than a hundred ITSM operations, a ceiling set per module, roles read from the instance and mapped; a three-step wizard.
HPE Morpheus
Several hundred cloud management operations in areas enabled separately for each assistant.
Veeam
Questions about backups in plain language, handled by Veeam Intelligence.
VMware vCenter
Virtual machines, hosts, clusters, datastores, snapshots, performance, events and alarms; read-only.
Any MCP server
Operation discovery, organisation and assistant ceiling, mode per operation, scope when connecting an account; use of the server's resources.
Connected accounts panel
The user connects and disconnects accounts and chooses the permission scope and the operations available to the assistant.
External sources and tools
Web search and page reading
Search with an adjustable number of results and retrieval of the content of an exactly specified page.
Lists of allowed and blocked addresses
The restrictions apply to every source invoked, together with the result scope and content volume.
Awareness of time
The current date and time in a set time zone, and date calculations.
More than 80 actions
Knowledge base, documents, materials, CRM, mail and calendar, Microsoft 365, external information, MCP and Skills.
Adaptive tool provisioning
The assistant receives the tools that fit the query and reaches for more during the work.
3.10 AI models and compliance · model catalogue from 3.14
AI models and compliance
The model is chosen per project. The catalogue shows ZDR and the EU region, and the requirements also apply to models working in the background: during indexing, summarisation and speech.
Available in version 1.7
6 functions · 2 newModels
Model catalogue and assignment to assistants
The catalogue is updated periodically and immediately after a change at the provider; per assistant: default model, name, modes, formats and parameters.
Central model configuration
A model seen from the organisation's side: in which assistants and with which parameters; assigning, detaching and changing a parameter in bulk.
Model catalogue with costs
An overview of models with costs, capabilities and compliance requirements, updated on demand.
Compliance
Zero Data Retention (ZDR)
Enforcing a mode in which the provider does not store content; models without ZDR are marked, and choosing a non-compliant one explicitly switches the requirement off.
ZDR and EU for auxiliary models
The requirements cover models that prepare queries and summaries, describe documents and spreadsheets and handle speech, with a filter for models that meet the requirement.
EU processing requirement
Processing in the European region; web search is blocked with an explanation when the model does not support it in this mode.
Development roadmap
3 in progress · 2 planned for developmentRoadmap items have no fixed date, and their scope may change.
LiteLLM model gateway
An intermediate layer between the platform and the model providers: routing, virtual keys, limits and billing, always in the path of traffic to the models. Module card 08.
Local model bridge
An adapter for OpenAI-compatible endpoints, e.g. vLLM, and a cost model for local models: amortisation of compute instead of a per-token rate.
Requires: LiteLLM model gateway
Local OCR on GPU
An OCR model in the customer's infrastructure, integrated with document processing and the large file sets module.
Local embedding
An embedding model for the knowledge base and the files module, run in the customer's infrastructure.
Requires: Local model bridge
Local reranking
A local relevance-scoring model in the search pipeline.
Requires: Local model bridge
3.11 Users, roles and permissions · 3.12 Authentication
Users, identity and audit
Security is about access: users from LDAP and Entra ID, roles configured in the UI from more than a hundred permissions, and a log that shows every change.
Available in version 1.7
10 functions · 4 newPermissions
Account management
Searching, creating, blocking, changing passwords, assigning to assistants; granting only roles within one's own permissions.
Roles defined by the organisation
Administrative permissions and permissions within an assistant down to a single field; menu filtered by permissions; default role.
Granular administrative permissions
Write access split into create, edit and delete; separate rights to grant roles, reset passwords and synchronise.
Permission audit
Who has access to what and on what basis, across people, roles, groups and assistants; filters for unused roles and inactive accounts.
User groups
Permissions granted to members in all assigned assistants; local groups or groups from Active Directory; delegated management.
Event log
Creations, changes and deletions with before and after values, logins and failed attempts, shares, system prompt changes; a technical account for tasks.
Identity
AD, Entra ID and local sign-in
A domain account verified by the domain controller, one-click Microsoft sign-in, a local account; separate paths.
Links that lead through sign-in
A link opened without signing in shows what will be displayed and leads there after sign-in; a preview with the name and logo.
Active Directory and Entra ID synchronisation
Connection test with diagnosis of causes, preview before import, directory structure, periodic synchronisation.
Entra ID synchronisation filter wizard
Account scope by department, job title, company, location, domain and account type, without Microsoft Graph query syntax.
3.13 Analytics, costs and limits · 3.14 Administration and white-label
Analytics, costs and administration
Every project has full billing, and the whole platform has a view one level up. You can see who used AI, when and at what cost, and limits keep the budget in check.
Available in version 1.7
12 functions · 1 newAnalytics and costs
Assistant analytics
Usage, cost, queries and users over time, by model, operation and person; the cost shown on every message.
Organisation and group analytics
The same views for the whole platform and for assistant groups; a deleted assistant stays in billing as an archived item.
Cost by operation type
Conversation, indexing, query preparation, OCR, spreadsheets, search, page retrieval, images, speech.
Limits and balance
A monthly budget or a prepaid package, a warning, blocking of queries once exhausted and a notice in the answer.
Spending limits for people and groups
A monthly limit for a user and a group, with a warning and blocking; the higher of the limits applies.
Conversation quality oversight
Review of an assistant's conversations with content and sources, within the granted permissions.
Activity: query log
Individual queries: who, when, which assistant, model and cost.
Administration
White-label
Your own brand mark in four variants: light mode, dark mode, logomark and browser tab icon.
Quality settings panel
Parameters for search, conversation, tools, materials, spreadsheets, speech, Deep Search, the recycle bin, the log, the sandbox and auxiliary models, with descriptions and dependency checks.
Background work and recurring jobs
Document processing, Skills, copying and synchronisation in the background with retries; periodic updates and cleanup.
Bilingual interface
A full Polish and English version, more than 1300 elements; the assistant answers in the user's language.
Assistant icon library
Nearly 900 symbols in three variants, expanded independently of application updates.
Development direction
Agents and tasks
The next step on the adoption curve: a workflow framework for multi-agent work and tasks that run on their own, within an agreed budget.
Development roadmap
3 in progress · 5 planned for developmentRoadmap items have no fixed date, and their scope may change.
Workflow framework
Orchestration on LangGraph: a supervisor and tool nodes, loops with an iteration limit, checkpointing, pausing for a human decision.
Agent graph interface
Building, previewing and debugging an agent flow.
Requires: Workflow framework
Flow tracing
Call traces, latency and the cost of every flow step.
Requires: Workflow framework
Task scheduling
Recurring tasks on a technical account, with the result sent to a designated recipient.
Webhook triggers
An incoming endpoint for invoking a task, deliberately opening inbound traffic to the installation.
Requires: Task scheduling
Task panel
Tasks per assistant, owner and schedule; run status, result preview, pausing, manual run, history.
Requires: Task scheduling
Task budgets
A cost limit per task and per assistant per month, a hard stop when exceeded, an alert at a threshold.
Requires: Task scheduling
Creating tasks from a conversation
The assistant proposes a schedule, a prompt and a result recipient for one-click approval.
Requires: Task scheduling, Task panel
Development direction · module card 09
Data protection (DLP)
A hardware and software component that censors data sent to external models. Personal data stays with the customer and the model works on pseudonyms.
Development roadmap
9 planned for developmentRoadmap items have no fixed date, and their scope may change.
DLP service
A local service called before every query to the model; fail closed, with redundancy.
Rule layer
Expressions with checksum validation for PESEL, NIP, REGON, Polish IBAN, ID card number, e-mail and phone; a Polish model for names and addresses; runs on CPU.
Requires: DLP service
Context layer
A local language model for indirectly identifying data and difficult cases.
Requires: DLP service, Local LLM for DLP
Local LLM for DLP
A language model in the customer's infrastructure, e.g. Bielik for the Polish language.
Requires: Local model bridge
Policy and decision
A deterministic decision to pass, mask or block according to the policy per assistant and data category.
Requires: DLP service
Pseudonym dictionary
A vault at the customer's site; values restored locally, including in tool arguments before an operation is executed.
Requires: DLP service
Data paths
The message to the model, tool and sandbox arguments and results, the answer to the user, and auxiliary models.
Requires: DLP service
Files
Redaction of images after OCR, permanent redaction of the PDF text layer, masking of Office files, metadata removal, preview before sending.
Requires: DLP service
Guarantees and audit
Hard guarantees for structured fields, measured effectiveness for free text, a checkpoint on the model gateway, DLP events in the log.
Requires: DLP service, LiteLLM model gateway
Development direction
Deployment and environment
The application is self-hosted. The target is a closed appliance: updates without access to external registries, data and backups in the customer's storage.
Development roadmap
6 planned for developmentRoadmap items have no fixed date, and their scope may change.
Immutable operating system
A system managed exclusively via an API with certificates, without SSH or a shell.
Secure Boot and disk encryption
Booting only a signed image; state and data partitions encrypted with a key sealed in the TPM.
Requires: Immutable operating system
Licence file
A signed file for the subscription period verified offline; after expiry a grace period, then read-only mode.
Offline updates
Signed release packages without access to external registries, with the option to roll back a version.
Requires: Immutable operating system
SBOM and log export
An SBOM for every release, export of logs and events to the customer's systems (syslog, OTEL).
Customer data outside the installation
Files and database backups in the customer's storage, restorable independently of the vendor.
Development roadmap
What we are building and what we are planning
Functions we are writing, and the directions in which we plan to develop the platform. For dependent items you can see what they build on.
In progress
9 itemsMobile app
Conversation, assistant selection, voice and artifacts on the phone.
Conversation with the assistantLarge file sets
Connecting shares and archives without copying: current documents with OCR and an index, the archive without vectorisation, cost at query time. Module card 05.
Requires: Local OCR on GPU, Local embedding
Knowledge and documentsChanges in documents
Editing documentation by conversation, version history, a file of the new version, questions about the change history. Module card 06.
Knowledge and documentsLiteLLM model gateway
An intermediate layer between the platform and the model providers: routing, virtual keys, limits and billing, always in the path of traffic to the models. Module card 08.
AI models and complianceLocal model bridge
An adapter for OpenAI-compatible endpoints, e.g. vLLM, and a cost model for local models: amortisation of compute instead of a per-token rate.
Requires: LiteLLM model gateway
AI models and complianceLocal OCR on GPU
An OCR model in the customer's infrastructure, integrated with document processing and the large file sets module.
AI models and complianceWorkflow framework
Orchestration on LangGraph: a supervisor and tool nodes, loops with an iteration limit, checkpointing, pausing for a human decision.
Agents and tasksAgent graph interface
Building, previewing and debugging an agent flow.
Requires: Workflow framework
Agents and tasksFlow tracing
Call traces, latency and the cost of every flow step.
Requires: Workflow framework
Agents and tasksPlanned for development
35 itemsRoadmap items have no fixed date, and their scope may change.
Very large PDFs
Streaming and paged processing in the background with progress, batch indexing, memory limits, first pages available while processing continues.
Knowledge and documentsSession sandbox
A persistent environment for the whole conversation with a durable file system, suspending and restoring state, CPU, RAM and disk limits; tools for working on files.
Files, artifacts and SkillsSandbox base image
Office and PDF tools, OCR with Polish language support, fonts with Polish characters; no network by default, protection against malicious files.
Files, artifacts and SkillsFile manifest in the prompt
The model knows the type, the number of pages, sheets or slides and a summary of the file, and reads the content on demand.
Requires: Session sandbox
Files, artifacts and SkillsFormats not supported by the model
Presentations, archives, binary files and legacy .doc, .xls, .ppt formats: conversion, extraction, ZIP archives with Polish file names.
Requires: Session sandbox, Sandbox base image
Files, artifacts and SkillsEditing Office files
Edits to DOCX, XLSX and PPTX in their own format, tracked changes, formula recalculation and error checks before the file is released.
Requires: Session sandbox, Sandbox base image
Files, artifacts and SkillsPDF conversions and operations
Office to PDF, DOCX and Markdown both ways, PDF to images, HTML to PDF; merging, splitting, rotating, forms, watermark.
Requires: Session sandbox, Sandbox base image
Files, artifacts and SkillsDrag-and-drop files
Several files at once onto the input field and the conversation window, with a preview before sending.
Files, artifacts and SkillsExtended artifact interface
Scope to be determined.
Files, artifacts and SkillsImport Skills from ZIP
Structure validation and registration in the assistant's catalogue after human approval.
Files, artifacts and SkillsSkills created by AI
A Skill's structure generated from a task description or materials, tested in the sandbox, saved with versioning after approval.
Requires: Session sandbox, Import Skills from ZIP
Files, artifacts and SkillsExplicit invocation of Skills
Asking for consent on a match, selection next to the chat field, forcing it in the conversation; mode per assistant: ask, automatic, manual only.
Files, artifacts and SkillsSkills catalogue interface
Sorting and filtering the catalogue; form to be determined.
Files, artifacts and SkillsLocal embedding
An embedding model for the knowledge base and the files module, run in the customer's infrastructure.
Requires: Local model bridge
AI models and complianceLocal reranking
A local relevance-scoring model in the search pipeline.
Requires: Local model bridge
AI models and complianceTask scheduling
Recurring tasks on a technical account, with the result sent to a designated recipient.
Agents and tasksWebhook triggers
An incoming endpoint for invoking a task, deliberately opening inbound traffic to the installation.
Requires: Task scheduling
Agents and tasksTask panel
Tasks per assistant, owner and schedule; run status, result preview, pausing, manual run, history.
Requires: Task scheduling
Agents and tasksTask budgets
A cost limit per task and per assistant per month, a hard stop when exceeded, an alert at a threshold.
Requires: Task scheduling
Agents and tasksCreating tasks from a conversation
The assistant proposes a schedule, a prompt and a result recipient for one-click approval.
Requires: Task scheduling, Task panel
Agents and tasksDLP service
A local service called before every query to the model; fail closed, with redundancy.
Data protection (DLP)Rule layer
Expressions with checksum validation for PESEL, NIP, REGON, Polish IBAN, ID card number, e-mail and phone; a Polish model for names and addresses; runs on CPU.
Requires: DLP service
Data protection (DLP)Context layer
A local language model for indirectly identifying data and difficult cases.
Requires: DLP service, Local LLM for DLP
Data protection (DLP)Local LLM for DLP
A language model in the customer's infrastructure, e.g. Bielik for the Polish language.
Requires: Local model bridge
Data protection (DLP)Policy and decision
A deterministic decision to pass, mask or block according to the policy per assistant and data category.
Requires: DLP service
Data protection (DLP)Pseudonym dictionary
A vault at the customer's site; values restored locally, including in tool arguments before an operation is executed.
Requires: DLP service
Data protection (DLP)Data paths
The message to the model, tool and sandbox arguments and results, the answer to the user, and auxiliary models.
Requires: DLP service
Data protection (DLP)Files
Redaction of images after OCR, permanent redaction of the PDF text layer, masking of Office files, metadata removal, preview before sending.
Requires: DLP service
Data protection (DLP)Guarantees and audit
Hard guarantees for structured fields, measured effectiveness for free text, a checkpoint on the model gateway, DLP events in the log.
Requires: DLP service, LiteLLM model gateway
Data protection (DLP)Immutable operating system
A system managed exclusively via an API with certificates, without SSH or a shell.
Deployment and environmentSecure Boot and disk encryption
Booting only a signed image; state and data partitions encrypted with a key sealed in the TPM.
Requires: Immutable operating system
Deployment and environmentLicence file
A signed file for the subscription period verified offline; after expiry a grace period, then read-only mode.
Deployment and environmentOffline updates
Signed release packages without access to external registries, with the option to roll back a version.
Requires: Immutable operating system
Deployment and environmentSBOM and log export
An SBOM for every release, export of logs and events to the customer's systems (syslog, OTEL).
Deployment and environmentCustomer data outside the installation
Files and database backups in the customer's storage, restorable independently of the vendor.
Deployment and environmentPLC ONE add-on modules
Add-on modules are ready-made use cases built on the platform. They are not part of the platform licence and are licensed separately. Each one uses what the platform already has: assistants, permissions, integrations and billing. If you need your own scenario, we have the developers and the know-how to build it with the customer.
10 add-on modules
AI Meeting Assistant
The team knows what it agreed.
Voice Tower · a PLC ONE module for meetings and project work
- A meeting processed a few minutes after it ends
- Conversations stay in the company
- Works with every meeting
Power BI integration
You talk to your reports as you would to an analyst.
Power BI · analytical conversation with the company's reports on the user's permissions
- Trends, relationships and causes
- Numbers consistent with the dashboard
- Everyone sees only their own data
Microsoft 365 Copilot integration
Copilot knows the whole company, and its licence works at full capacity.
Microsoft 365 Copilot · more from the licence and access to data from the whole company
- Copilot as a token factory
- Copilot sees ERP, CRM and the company's systems
- Licences where they are needed
Knowledge from manuals and screenshots
A solution in seconds, even if it was recorded in a screenshot.
OneNote and other applications · image and text read together as one instruction
- A solution instead of searching
- Image and text read together
- Not just OneNote
Working with large file sets
A conversation with the company's entire archive, without sending it to the cloud.
Company files in conversation with AI · beyond the limits of RAG, without transfer to the cloud
- Beyond the limits of RAG
- Files stay private, also in air-gap mode
- High quality at low cost
Documentation change register
The document changes in a conversation with AI, with a version register.
Technical documentation, procedures and ISO documents · editing with AI, version history, a file of the new version
- Editing documents with AI
- Versions and change history
- A file of the new version after the changes
IT integration pack
The assistant analyses the company's IT systems based on the ManageEngine CMDB.
ServiceDesk Plus and CMDB, Jira, HPE VM Essentials, Veeam · IT systems analysis through the CMDB
- Systems analysis through the CMDB
- IT system operations in natural language
- Permissions from source systems
LiteLLM model gateway
Every model, including local or your own, connects without changes to the application.
Cloud, local and the customer's own models under one interface · routing, limits, billing
- One interface for all models
- Local models in the platform catalogue
- Costs and limits in one billing
DLP module
Personal data stays in the company, even when working with models in the cloud.
Local detection and pseudonymisation of personal data · an appliance prepared for the customer
- Detection on a local model
- Reversible pseudonymisation
- DLP where it is needed
E-commerce module
The offer is built from the ERP product record, and AI fills in the missing data.
From the ERP product record to a published offer · templates, pre-submission checks, AI at every stage of the process
- Product data from documents, not just from fields
- Mapping and content for every platform
- Pre-submission check and publication register
AI Meeting Assistant
Voice Tower · a PLC ONE module for meetings and project work
A PLC ONE module that turns meetings into the current state of a project's decisions. The transcript, outline, decisions and tasks are ready a few minutes after the conversation. They stay in the company's environment, in line with its data protection policy.
Main value
The team knows what it agreed.
The module keeps one register of a project's binding decisions and updates it after every meeting. When a new meeting contradicts an earlier decision, the system shows both sources and proposes a change. The change takes effect only after a human accepts it.
A meeting processed a few minutes after it ends
A transcript split by speaker, an outline, three summaries for different audiences and a task list with deadlines. Nobody has to start anything.
Conversations stay in the company
Transcripts and notes are stored in PLC ONE, in line with the company's privacy policy. The voice recording can be deleted right after transcription.
Works with every meeting
Teams, Zoom and Google Meet. A meeting in a room is recorded by a phone placed on the table, even without signal.
The work cycle: from meeting to meeting
Every meeting goes through six steps. The sixth compares the new decisions with the register and opens the next cycle.
- 1Meeting
Retrieved from Teams, Zoom or Meet, by a bot or the mobile app.
- 2Transcription and processing
Transcript, outline, three summaries, tasks. The recording can be deleted.
- 3Completion
The author's context, speaker assignment, a note in draft status.
- 4Approval
Tasks to the CRM, confirmation by e-mail, entry into the project knowledge.
- 5Work between meetings
Questions to the repository, an answer with a quote from the transcript.
- 6The next meeting
Comparison with the decision register; on a conflict, a proposed change.
After the last step the cycle starts again.
Recording
Transcription
Processing
Project knowledge
Decisions
Output
Control
Principle
Integrations
Microsoft Teams, Zoom, Google Meet, Microsoft 365, Zoho, Jira, ServiceDesk
Uses from the platform
Chat with knowledge, permissions, document versioning, integrations
Power BI integration
Power BI · analytical conversation with the company's reports on the user's permissions
A PLC ONE module that turns Power BI reports into a partner for analytical conversation. The assistant answers questions about the data, compares periods, points out trends and relationships, and proposes conclusions. It takes the numbers from the models visible on the dashboard, on the permissions of the person asking.
Main value
You talk to your reports as you would to an analyst.
The assistant understands the question and its context, builds the queries to the semantic models itself, compares periods, looks for trends and relationships between measures, and finally draws conclusions. Each follow-up question refines the previous one, as in a conversation with an analyst who knows all the company's reports.
Trends, relationships and causes
The assistant compares periods, picks up deviations, calculates correlations between measures and explains what has changed. The question “why” is the start of the analysis, not the end of the conversation.
Numbers consistent with the dashboard
The data comes from Power BI semantic models, with the report and refresh time indicated. Whatever the assistant calculated or inferred on its own is marked as such.
Everyone sees only their own data
The query runs in Power BI on the permissions of the signed-in user, including row-level security. PLC ONE does not keep a second list of permissions.
The path of a question: from conversation to conclusion
Every question goes through five steps. The data always comes from Power BI semantic models, on the permissions of the person asking; the analysis and conclusions are the assistant's work.
- 1Question and context
Natural language, the context of earlier questions in the conversation.
- 2Report selection
Catalogue and scope rules; on conflicting definitions, a follow-up question.
- 3Queries on behalf of the user
The assistant builds DAX queries; on-behalf-of, RLS from Power BI.
- 4Analysis
Period comparisons, trend, relationships, calculations in the sandbox.
- 5Answer with a conclusion
Numbers with the source and refresh time, interpretation marked.
Report catalogue
Semantic dictionary
Access and permissions
Report selection
Analysis and conclusions
Conversation
Maintenance and audit
Principle
Integrations
Power BI (semantic models, DAX queries), Microsoft Entra ID
Uses from the platform
Company sign-in, computational sandbox, event log, versioning
Requirements
A Power BI Pro licence for the people asking (in Microsoft 365 E5) or a workspace on F64+ capacity
Microsoft 365 Copilot integration
Microsoft 365 Copilot · more from the licence and access to data from the whole company
A PLC ONE module for companies that have Microsoft 365 and Copilot. It directs to Copilot the work the company has already paid for in the licence, and extends it with data from ERP, CRM and the company's other systems. The integration works both ways, and the company decides what from its business systems reaches Microsoft.
Main value
Copilot knows the whole company, and its licence works at full capacity.
Questions about mail, meetings and documents are handled by Copilot within the licence, and questions about business data by PLC ONE models. The user gets one answer, in PLC ONE or in Copilot, with data from Microsoft 365 and from the company's systems. The company pays less for models and makes better use of the licences it already has.
Copilot as a token factory
A Copilot licence is AI work paid for in advance. PLC ONE directs summaries of threads, meetings and documents to it, and the statistics show how many platform tokens were saved as a result.
Copilot sees ERP, CRM and the company's systems
Through PLC ONE, Copilot gets data from CRM, ServiceDesk, Jira, Power BI and any MCP-compatible system. The company does not build separate Copilot extensions for each system.
Licences where they are needed
The model is chosen for each user according to their licence. Copilot for people who work intensively in Microsoft 365, PLC ONE models for everyone else, with access to mail, calendar and Teams.
Integration in both directions
A question can be asked in PLC ONE or in Copilot. The assistant routes it to where the answer is, and only what the company's policy allows passes between Microsoft 365 and the company's systems.
- 1Question
In PLC ONE, in Copilot or in Teams; the conversation context and the user's licence.
- 2Routing
Mail, meetings, documents to Copilot; business data to the platform's models.
- 3Microsoft 365 track
Copilot within the licence: semantic index, synthesis of mail, Teams and meetings.
- 4Company data track
PLC ONE tools: ERP, CRM, ServiceDesk; permitted fields, masking.
- 5Answer
One answer with sources; an entry in the transfer register and statistics.
Copilot in PLC ONE
PLC ONE for Copilot
Token factory
Model and licence selection
Document search
Transfer control
Confidentiality and register
Principle
Integrations
Microsoft 365 Copilot, SharePoint, OneDrive, Teams, Microsoft Entra ID
Uses from the platform
M365, CRM, ServiceDesk and MCP integrations, model catalogue, sandbox, event log
Requirements
Microsoft 365 Copilot licences for the users who are to use it
Knowledge from manuals and screenshots
OneNote and other applications · image and text read together as one instruction
A PLC ONE module for applications in which the company keeps instructions made of text and screenshots. A vision model combines the image with the text according to a description instruction, and the assistant answers from the knowledge prepared this way in seconds.
Main value
A solution in seconds, even if it was recorded in a screenshot.
Instead of browsing notebooks and instruction pages, an employee describes the problem in their own words. The assistant searches all the connected knowledge, including the content of screenshots. It answers with concrete steps, indicating the page and image the answer comes from.
A solution instead of searching
A question in plain language replaces browsing hundreds of pages. The answer leads step by step and links to the original page, so you can see the screenshot with your own eyes.
Image and text read together
The vision model recognises the system window, highlighted fields, arrows and error messages in a screenshot. The description instruction decides how the image is combined with the text, so the step “click here” gets the information where “here” is.
Not just OneNote
OneNote is the first application in which the image is combined with the text by means of a description instruction. PLC ONE applies the same mechanism to other applications with instructions, building a dedicated module for them.
From instructions with images to an answer
Every instruction page goes through five steps, regardless of the application it was created in. Synchronisation repeats them for every changed page.
- 1Connecting the source
OneNote or another application; the administrator chooses the scope.
- 2Reading the page
Text, tables and images in the order in which they appear.
- 3Combining image and text
The vision model according to the description instruction, in the chosen language.
- 4Knowledge base
A description in place of the image; the instruction keeps the order of the steps.
- 5Answer with the source
Steps to perform, the page and image, a link to the original.
After the last step the cycle starts again.
Vision model
Description instruction
Language and translation
Reading the page
OneNote connector
Other applications
Synchronisation
Principle
Integrations
Microsoft OneNote (Microsoft Graph) as the first application; further ones on the same mechanism
Uses from the platform
Knowledge base and search, model catalogue (vision model), OCR, assistants
Requirements for OneNote
Notebooks in OneDrive or SharePoint, materials in sections without password protection
Working with large file sets
Company files in conversation with AI · beyond the limits of RAG, without transfer to the cloud
A PLC ONE module that goes beyond the limits of classic RAG and lets you converse with large sets of company files. With local AI models and the PLC ONE platform in the company's infrastructure, air-gapped operation is possible.
Main value
A conversation with the company's entire archive, without sending it to the cloud.
Classic RAG works well on a few dozen uploaded files, but it cannot cope with large sets in which every folder holds hundreds of files. The module combines three ways of working with files: virtual folders in the project's knowledge base, synchronisation of external folders into RAG with full OCR, and indexing of large sets, after which the assistant works directly with the source.
Beyond the limits of RAG
There is no need to upload files or process the whole archive up front. Current documents get full preparation, the archive is available from day one, and exact search complements semantic search where numbers, amounts and names matter.
Files stay private, also in air-gap mode
With local AI models and the PLC ONE platform in the company's infrastructure, work with files happens in a closed loop. The company does not need a data processing agreement with the model provider.
High quality at low cost
The assistant combines OCR, semantic and exact search, independent document search and the sandbox. The company pays for the archive only at query time, and only for the files selected for the answer.
Three ways of working with files
For each external folder the company chooses the integration method: synchronisation into RAG or indexing. All methods work within the project and the user's permissions.
- 1Connecting sources
SMB/NFS network shares and SharePoint.
- 2Integration method
Synchronisation into the project's RAG with full OCR, or indexing of a large set.
- 3Question and search
Virtual folders, or the index and reading the source; exact search, semantic search, navigation.
- 4Answer from the source
Content from the documents, file name, path and fragment.
Virtual folders
Synchronisation into RAG
Indexing large sets
Choice of integration
Exact search
Navigation
Air gap
Principle
Integrations
SMB/NFS network shares (Windows file servers), SharePoint
Uses from the platform
Knowledge base, OCR, semantic search, independent document search, sandbox, model catalogue (including local models)
Requirements
Read access to the connected resources; connect scans without a text layer through synchronisation with OCR
Documentation change register
Technical documentation, procedures and ISO documents · editing with AI, version history, a file of the new version
A PLC ONE module for working with technical documentation, procedures and management system documents, e.g. ISO. A document is corrected, edited and changed in a conversation with AI, and the module keeps a version register.
Main value
The document changes in a conversation with AI, with a version register.
The user describes what should change. The AI finds the place in the document, edits the content in its format and shows a preview of the changed fragments. After approval the module saves a new version with an automatic description of the change.
Editing documents with AI
The AI makes changes described in a sentence, rewrites fragments on request and keeps the document's conventions, e.g. adds a row to an existing table in the same format.
Versions and change history
Every save creates a new version, and previous ones are kept in full. The module shows the author, date and description of the change, compares versions, restores the state on a chosen day and lets you undo a change.
A file of the new version after the changes
When the work is finished, the module generates the document file in its current version. The module therefore serves not only for keeping documentation in PLC ONE, but also for working on documents with AI that ends with a new file.
From instruction to a file of the new version
Every change goes through five steps. Saving happens after the user's approval, and a file of the new version can be generated once work on the document is finished.
- 1Instruction
A description of the change or an editing instruction in natural language.
- 2Editing by AI
Finding the place, a change in the document's format.
- 3Preview
Changed fragments in context, a summary, a description for the history.
- 4Approval and version
Saving a new version or correcting the proposal.
- 5File of the new version
Generating the document file after the changes.
Editing with AI
Preview and approval
Versioning and change descriptions
Questions about history
File of the new version
Documentation screen
Teamwork
Principle
Integrations
Requires no external integrations; runs on the PLC ONE platform
Uses from the platform
Assistants and AI models, permissions, full-text search, event log; the screen linked with the chat both ways
Requirements
A PLC ONE project
IT integration pack
ServiceDesk Plus and CMDB, Jira, HPE VM Essentials, Veeam · IT systems analysis through the CMDB
A PLC ONE pack for the IT department, made up of existing platform integrations. Its central element is the integration with the ManageEngine CMDB, which serves as an intermediate layer for examining and analysing the company's IT systems.
Main value
The assistant analyses the company's IT systems based on the ManageEngine CMDB.
The CMDB contains configuration items, their relationships, assets and related tickets, problems and changes. The assistant uses it as an intermediate layer: it answers questions about dependencies between systems, the impact of changes and the causes of failures, and prepares analyses.
Systems analysis through the CMDB
The assistant links configuration items and their relationships with tickets, problems and changes, and answers analytical questions, e.g. what depends on a given server or which changes preceded a failure.
IT system operations in natural language
The ticket lifecycle in ServiceDesk Plus, issues in Jira, operations in HPE VM Essentials and backup status in Veeam. For every operation the administrator decides whether it is allowed, requires approval or is blocked.
Permissions from source systems
User roles are read from ServiceDesk Plus and refreshed daily, and an unmapped role means no permissions. The event log records every operation with the value before and after the change.
Analysis through the CMDB, operations through integrations
The assistant routes questions about IT systems and their dependencies to the ManageEngine CMDB. It performs operations and status checks through integrations with the individual systems, within the administrator's ceiling and the user's permissions.
- 1Question or instruction
Natural language, the conversation context and the user's permissions.
- 2ManageEngine CMDB
Configuration items, relationships, assets, related tickets, problems and changes.
- 3Operational integrations
ServiceDesk Plus, Jira, HPE VM Essentials, Veeam; operation mode set by the administrator.
- 4Answer or operation
An analysis with links to objects, or a completed operation, with an entry in the event log.
ManageEngine CMDB
Analysis and analytics
ServiceDesk Plus
Jira, Confluence, Compass
HPE VM Essentials
Veeam
Other systems
Permissions and log
Integrations
ManageEngine ServiceDesk Plus (with CMDB), Jira, Confluence, Compass, HPE VM Essentials (Morpheus), Veeam
Uses from the platform
Integrations and MCP, permissions, operation approval, event log, assistants
Requirements
A CMDB maintained in ManageEngine; API access to the connected systems; accounts and roles in ServiceDesk Plus
LiteLLM model gateway
Cloud, local and the customer's own models under one interface · routing, limits, billing
A PLC ONE module that introduces LiteLLM as an intermediary layer between the platform and AI model providers. Cloud models, local models (vLLM, Ollama, HPE Private Cloud AI) and the customer's own models are available through one OpenAI-compatible interface, without changes to the application.
Main value
Every model, including local or your own, connects without changes to the application.
A model is added in the settings panel with an address and a key. It appears in the platform's model catalogue alongside cloud models and can be assigned to assistants. For every model the gateway handles routing, failover to a backup model, query retries and limits.
One interface for all models
PLC ONE assistants use cloud, local and own models in the same way. Changing the model in an assistant or adding a new model does not require changes to the application, so the company is not tied to a single provider.
Local models in the platform catalogue
Models running in the company's infrastructure, on vLLM, Ollama or in HPE Private Cloud AI, are available in PLC ONE like any other model. Assistants working on confidential data can be assigned to local models only.
Costs and limits in one billing
The gateway counts tokens and costs for all models consistently and passes them to PLC ONE billing: per message, user and assistant. Query limits are set separately for each model.
The path of a query through the model gateway
Every assistant query passes through the LiteLLM gateway. The gateway routes it to the model indicated in the PLC ONE catalogue, enforces the limits and returns usage data to the platform's billing.
- 1Assistant query
The model assigned in the PLC ONE catalogue.
- 2LiteLLM gateway
One OpenAI-compatible interface, a key and a limit for the model.
- 3Routing
Target model, retries, failover to a backup model.
- 4Model
Cloud, local (vLLM, Ollama, HPE Private Cloud AI) or the customer's own model.
- 5Billing
Tokens and cost to PLC ONE analytics and limits.
One interface
Local models
Customer's own models
Routing and resilience
Limits
Costs
PLC ONE model catalogue
Principle
Integrations
LiteLLM; vLLM, Ollama, HPE Private Cloud AI; providers and brokers of cloud models
Uses from the platform
Model catalogue, settings panel, cost analytics and limits, assistants
Requirements
A LiteLLM instance in the company's infrastructure or in HPE Private Cloud AI; GPU infrastructure for local models
DLP module
Local detection and pseudonymisation of personal data · an appliance prepared for the customer
A PLC ONE module that detects personal and confidential data before a query is sent to the AI model, and masks or blocks it according to the company's policy. Detection is based on a local model and runs entirely in the customer's infrastructure.
Main value
Personal data stays in the company, even when working with models in the cloud.
Projects with DLP enabled pass through a local control service before every query to the model. Personal data is replaced with pseudonyms or the query is blocked. The pseudonym dictionary stays with the customer, and in the answer to the user the real values are restored locally.
Detection on a local model
A local model classifies content and documents for personal and confidential data and decides on the response: pass, mask or block. It also recognises data that identifies a person indirectly.
Reversible pseudonymisation
The AI model receives pseudonyms instead of personal data, and the values return to the answer locally. The company can therefore use cloud models when working with documents that contain personal data.
DLP where it is needed
The module is enabled for selected projects, e.g. HR or finance, while the platform's other projects run without the additional control and without its cost. The rules and data categories are parameterised to the customer's requirements.
Data control before sending to the model
In projects covered by DLP, every query, attachment and tool result passes through the local DLP service. The model receives content that complies with the company's policy, and the real values return to the answer locally.
- 1Project covered by DLP
A message, an attachment, arguments or a tool result.
- 2DLP service on the appliance
A local model and a rule layer.
- 3Response according to policy
Pass, pseudonymisation or block.
- 4AI model
Content with pseudonyms; independent control on the model gateway.
- 5Answer
Values restored locally from the pseudonym dictionary.
Local DLP model
Presidio, a complementary layer
Pseudonymisation
Data paths
Attachments
Policy and scope
Appliance
Guarantees and audit
Integrations
Local DLP model, Presidio with the Polish spaCy model, Tesseract OCR, PyMuPDF
Uses from the platform
Projects, model gateway, event log, analytics, assistant policies
Requirements
A DLP appliance delivered with the module; the projects in which DLP is to run
E-commerce module
From the ERP product record to a published offer · templates, pre-submission checks, AI at every stage of the process
A PLC ONE module for selling in e-commerce channels. Product data is pulled from the ERP system, assembled into an offer according to a template assigned to the category and platform, and published after a preview and human acceptance.
Main value
The offer is built from the ERP product record, and AI fills in the missing data.
The operator selects the products, and the module assembles the offer according to a template, checks for gaps before submission and shows a preview. The AI fills in the missing data based on supplier documents, matches it to the platform's requirements, prepares the content and checks the offer before it is sent.
Product data from documents, not just from fields
The AI reads supplier datasheets, manuals and specifications and proposes values for missing attributes, indicating the document and the place they come from. It detects contradictions between sources and groups product variants into one offer.
Mapping and content for every platform
The AI proposes a category in the platform's tree, matches attributes to its dictionaries and prepares titles, descriptions and language versions in the brand's style. Dimensions, standards and certificates are carried over from the product record, not generated by the model.
Pre-submission check and publication register
Products that cannot be published go onto a gap list with a specific reason. For every offer it is recorded who published it and when, from which template and on which version of the data.
From the product record to a published offer
The operator works on products and templates. Every offer goes through five steps, and the AI supports each of them with proposals for acceptance.
- 1Product selection
One by one or as a set, by category, brand, SKU, from the ERP product record.
- 2Template and offer
Category and platform template; title, parameters, description, images, category.
- 3Pre-submission check
A gap list with the reason; quality control by AI.
- 4Preview and acceptance
The offer in a form close to the final one; nothing goes out without acceptance.
- 5Publication and register
Who, when, from which template, on which version of the data.
ERP integration
Templates
Sales platforms
Batch work
AI: product data
AI: mapping and content
AI: image and quality control
Principle
Integrations
The ERP system; sales platforms, each as a separate scope
Uses from the platform
Company sign-in, permissions, event log, budget limits, cost report, work with documents
Requirements
Access to the ERP product record; accounts on the sales platforms; configuration separately for each project