Connect your data sources
Documents, knowledge bases and corporate data sources are securely integrated into the system.
PrivateGPT is a secure AI assistant that works with internal documents, corporate data and knowledge bases. It runs in two modes: in local mode data never leaves your infrastructure, and in external-model mode you connect your own ChatGPT, Claude, Gemini, Grok or DeepSeek key while the uploaded file and the prompt you write are anonymized. It is KVKK compliant and offers on-premise or private cloud deployment.
Is there a penalty clause over [AMOUNT_1] in the agreement signed with [PERSON_1]?
sent to model · maskedThe agreement does include a penalty clause. In case of late delivery, a daily penalty of two per mille is applied over the [AMOUNT_1] amount. The cap is ten percent of the contract value.
Message PrivateGPT
Redactra protection is on; personal data is never sent unmasked.
Many AI platforms may use customer data for model training or service development, which creates serious privacy and compliance risks for organizations. PrivateGPT offers an architecture that keeps control of corporate data entirely inside the organization. When the power of external models is needed, data is not sent as-is; personal data is anonymized in the Redactra layer. This way you benefit from AI capabilities while data security and corporate confidentiality are preserved.
Data never crosses this boundary
Anonymized content only
Organizations can work at two different privacy levels on one platform. Alongside the local model, you can connect ChatGPT, Claude, Gemini, Grok and DeepSeek using your own API key. Users decide which mode each conversation runs in, and the interface always shows the active mode.
The model runs on your own infrastructure. Conversations, uploaded documents and generated responses stay within system boundaries; no connection is made to any external service. It also runs on air-gapped networks without internet access, making it suitable for the highest-security environments.
This is not an off-the-shelf model used as-is. It runs on our own layer, trained for corporate document language.
Who it is for: Banks, public institutions, defense and healthcare organizations where data is never permitted to leave the organization under any circumstances.
You connect your own API key. Before submission, both the uploaded file content and the prompt you write pass through the Redactra layer, where personal data is masked. The model provider never sees real personal or corporate information.
All are used with your own API key, through the Redactra layer.
Who it is for: Legal, finance and consulting teams that need the capabilities of powerful external models but do not want to expose client and employee data.
When the working mode changes, the interface, document management and chat experience stay the same. The only thing that changes is which model runs the work and how far the data travels.
PrivateGPT provides a secure AI infrastructure that works without sending corporate data to external platforms. All data processing runs under the organization's control.
By preventing corporate data from being passed as-is to third-party models, it provides a usage model compliant with KVKK and corporate data policies.
InfinityQ Model runs inside the organization. If you wish, you connect your ChatGPT, Claude, Gemini, Grok or DeepSeek key; because the content sent is anonymized, these models never see real corporate data.
Most of the risk lies not in the document but in the sentence the user types into the chat. PrivateGPT masks both channels in the same layer.
Organization-specific AI assistants can be built on corporate documents, knowledge bases and operational data.
No document you upload is used for model training. This commitment applies to both working modes.
Where the infrastructure is located is decided together, in line with the organization's security policies and regulatory requirements.
PrivateGPT runs on the organization's own servers. Data flow stays entirely within the corporate infrastructure; no external connection is required.
Running in a private cloud environment, an infrastructure compliant with corporate security policies is set up. No hardware investment is required.
Once deployment is complete, usage proceeds in three steps. At every stage the user sees how far the data travels.
Documents, knowledge bases and corporate data sources are securely integrated into the system.
Per conversation, either the local model or an external model you have connected is used. In external mode, content is anonymized before it is sent.
Users query corporate data in natural language and receive source-grounded answers.
It can be used in any process where corporate knowledge is scattered and access takes time.
Enables fast search and analysis across contracts, regulatory texts and legal documents.
Creates a smart internal knowledge assistant working on company documents and procedures.
Support assistants can be built on internal documentation and operational knowledge.
Provides fast information access and contextual analysis across large document collections.
In enterprise AI usage the risk is often not in the uploaded document but in the sentence the user types into the chat. PrivateGPT handles both channels in the same layer.
File content and prompt text are scanned together. Person names, ID numbers, address and contact details, account and file numbers, and company titles are identified.
Identified data is replaced with consistent labels. The same person gets the same label everywhere in the text, so meaning and the model's reasoning stay intact.
Before the external model's answer is shown to the user, labels are matched back to their real values. The mapping table stays within the organization's boundary.
Pricing
Local usage is unlimited in every plan and does not count against any allowance. In plans that use external models, documents are anonymized by the Redactra layer before they are sent, and only those requests count. There is a separate plan for users who never send documents outside.
All features · 10 queries and 3 documents to external models
No credit card required. Everything including the Redactra layer can be tested; only the volume is limited.
For users whose documents must never leave the organization under any circumstances, working with the local model only.
Chat and document analysis are unlimited. There is no usage limit on this plan.
No document is ever sent outside. External model connectivity is not available in this plan.
Unlimited chat with the local model
Document analysis with the local model
Documents are never used for training
Connect your own API key
Redactra anonymization layer
REST API
For those who want to use ChatGPT, Claude, Gemini, Grok and DeepSeek models under Redactra protection.
TRY 19.90 per document · excl. VAT
Chat and local document analysis are unlimited. No allowance is used as long as you use the local model.
Only outbound documents are counted. Documents sent to external models pass through the Redactra layer and personal data is anonymized.
Connect your own API key
Per-conversation model selection
Redactra anonymization layer
Documents are never used for training
REST API
On-premise deployment
For teams, regulated organizations and any structure where data must remain inside the company.
Based on user count and deployment
No quota applies. Capacity is determined by your own hardware.
Every document sent to an external model is anonymized. No document quota applies.
Everything in Individual
On-premise or private cloud deployment
Role-based access and audit trail
REST API and system integration
Deployment, training and technical support
Service level agreement
All prices exclude VAT. Chat and document analysis performed on the local model never count against your quota; the monthly quota applies only to documents sent to external models through the Redactra layer. External models are served through limited-use keys provided by InfinityQ; you may also connect your own provider key. Your own keys are kept only in your browser and are never stored on our servers. Top-up packs are granted only to active subscribers and are added to the subscription term; unused allowance does not carry over once the subscription ends. Fair use and document retention terms apply to local usage.
Frequently asked questions
Corporate data stays inside the organization; when external models are needed, it is anonymized before it is sent. Employees benefit from powerful AI capabilities without compromising data security.