Skip to main content

ZCore

Private AI Infrastructure

ZCore is your private AI engine: open-source language models on dedicated EU infrastructure. Chat, document intelligence, transcription and multimodal AI - via API or ready-made apps. Data never leaves the EU and is never used for training.

Private language model

Open-source models you control. Your data never trains anyone else's model.

Multimodal processing

Text, images, PDFs, Word files and audio in one engine - ask questions across all of them.

API-first architecture

Ready for your stack: APIs and integrations to CRM, ERP and document systems - build in-house or with a partner.

Lightning-fast inference

Dedicated European infrastructure means fast answers for every user.

Read more

ZCore

Private AI Infrastructure. Built for modern organisations.

ZCore is Zeldoc's private AI engine, built on specialised components - not one monolithic model. It handles conversation, coding, vision, voice, document processing and web search, all on dedicated EU infrastructure.

Under the hood, ZCore consists of a series of underlying models, each handling specific tasks within the language engine. Together they appear to the user as one seamless, private AI assistant.

Each component is designed for a concrete user-facing capability. ZCore handles text analysis, conversation and multilingual language understanding. ZTask provides coding intelligence, technical reasoning and tool-based actions such as web search. ZTranscribe transcribes meetings, notes and audio into text. ZEmbed enables retrieval-augmented generation (RAG) by embedding uploaded documents for AI-powered search and analysis.

Organisations can use ZCore as the foundation for private, business-secure AI applications: internal AI assistants, knowledge platforms, document intelligence, meeting transcription and business workflows. Data stays fully under control - everything is processed on dedicated European infrastructure without third parties gaining access to sensitive information.

ZCore delivers GDPR compliance, AI Act readiness and Zero Data Retention on local/EU models.

What ZCore does not do

ZCore does not include local models for image generation or video generation. These capabilities require external frontier models accessed through ZRouter, which means data leaves the EU at inference.

Highlights

  • Private LLM on dedicated EU infrastructure
  • Multimodal: text, images, files, documents, audio
  • Document intelligence & meeting transcription
  • API-first: CRM, ERP, documents, databases
  • Zero Data Retention - no third-party access
  • GDPR & AI Act ready

What powers ZCore

Specialised AI components, not one model

Each component below handles one concrete capability. Local components run entirely on EU infrastructure - image and video generation require external models.

ZCore

Text analysis, conversation and multilingual language understanding. Powers chat, summarisation, rewriting and reasoning across languages.

ZTask

Coding intelligence and technical reasoning for software development workflows, code generation, technical problem-solving and tool-based actions such as web search.

ZTranscribe

Transcribes meetings, notes and audio into text with high accuracy across multiple languages (ASR).

ZEmbed

Embeds uploaded documents for retrieval-augmented generation (RAG), enabling AI-powered search and analysis of your knowledge base.

Image & video generation

Not available locally. Requires external frontier models via ZRouter - data leaves the EU at inference and Zero Data Retention cannot be guaranteed.

Quality benchmark

ZCore vs. leading frontier models

Quality grades for concrete, user-facing capabilities. Grades are on a 1-5 scale with one decimal. Higher is better.

CapabilityZCore 1.1 (previous model)ZCore 2.0Sonnet 4.6Sonnet 5.0Opus 4.8Opus 5.0ChatGPT 5.6 SolGemini 3.5
Text analysis & conversation4.24.64.14.14.64.74.73.8
Coding & technical tasks4.54.64.44.54.64.74.74.1
Read PDF documents3.93.94.54.64.74.84.44.6
Read Word documents3.93.94.44.54.64.74.34.3
Read PowerPoint†3.53.54.34.44.54.64.24.2
Read Excel†3.53.54.24.34.44.54.54.3
Image reading & understanding3.83.84.44.54.64.74.54.7
Image generation------4.74.5
Video generation------4.03.8
Web search3.73.74.34.44.54.64.64.7
Task execution (agentic)3.93.94.44.54.64.74.54.2
Text to voice (TTS)4.04.0----4.34.0
Voice to text (ASR)4.34.3----4.34.0
Multilingual handling4.04.04.54.64.74.84.54.6
Document understanding (RAG)4.14.1----4.24.1
1-5 scale with one decimal. Higher is better.- = capability not available for this modelZCore 1.1 previous model - grades reflect proprietary, EU-hosted componentsZCore 2.0 based on next-gen architectureExternal models graded as used directly (US-hosted) PowerPoint and Excel reading in ZCore requires a partner extension. Grades are based on Zeldoc's internal testing and public benchmarks. External model grades reflect quality only - see the data sovereignty table for where data is processed.

Data sovereignty & GDPR

Where does your data go?

For each capability, this table shows whether data stays in the EU (data sovereign) and whether the function is GDPR compliant - for ZCore and for external frontier models.

CapabilityZCore: Data sovereignZCore: GDPRExternal frontier models
Text analysis & conversationYesYesNot sovereign
Coding & technical tasksYesYesNot sovereign
Read PDF documentsYesYesNot sovereign
Read Word documentsYesYesNot sovereign
Read PowerPoint†YesYesNot sovereign
Read Excel†YesYesNot sovereign
Image reading & understandingYesYesNot sovereign
Image generationNo - requires externalPartialNot sovereign
Video generationNo - requires externalPartialNot sovereign
Web searchNo - data sent to search enginesPartialNot sovereign
Task execution (agentic)YesYesNot sovereign
Text to voice (TTS)YesYesNot sovereign
Voice to text (ASR)YesYesNot sovereign
Multilingual handlingYesYesNot sovereign
Document understanding (RAG)YesYesNot sovereign
ZCore local components (ZCore, ZTask, ZTranscribe, ZEmbed): 100% data sovereign - data stays in the EU, no third-country transfer, no US cloud dependency, no FISA/702 exposure. Web search (tool call via ZTask): search queries are sent to external search engines, so this feature is not data sovereign - but it can be switched off. Image & video generation: no local ZCore model - requires external frontier models via ZRouter; data leaves the EU at inference. All external frontier models (Claude Sonnet 4.6, Claude Sonnet 5.0, Claude Opus 4.8, Claude Opus 5.0, ChatGPT 5.6 Sol, Gemini 3.5): US-based; data leaves the EU at inference. ZCore 1.1 and ZCore 2.0 run on dedicated European infrastructure with Zero Data Retention. GDPR compliance is possible through Standard Contractual Clauses (SCCs) and the EU-US Data Privacy Framework, but data does not stay in the EU. The customer is responsible for ensuring a legal basis for the transfer under GDPR Art. 44-49. See Privacy Policy for details.

Key takeaways

What this means for your organisation

Competitive quality, sovereign by design

Strong AI performance across text, coding, documents, vision and voice - with 100% EU data sovereignty for all local components and no third-party access to your information.

Honest about the gaps

Image and video generation require external models via ZRouter, so data leaves the EU for those calls. Web search is not data sovereign either - queries are sent to external search engines.

The sovereignty advantage

External models may score higher on some benchmarks, but ZCore keeps your data in the EU - for 12 out of 15 capabilities. For the remaining 3, ZRouter provides governed access with policy control through ZControl.

Explore ZCore

See pricing, request test access or talk to us about your AI requirements.