Fully local processing
Operational content stays within your network when all required models, embeddings, document and supporting services, and the sources in use, run locally. We validate the scope before rollout.
On-Premise · Private AI · Data Sovereignty
Your data.
Under your control.
Self-hosted on your infrastructure, including for small and midsized businesses. For Private AI, we plan supported local AI models and every processing service required for your tasks together.
chifty / SECURITYInteractive illustration
Access starts with you. Identity and permissions.
Local data flows
A data path you control.
The local path connects approved sources, permission checks, knowledge retrieval and supported local AI. Cloud sources and email providers remain external systems. Model endpoints, embeddings, document processing and OCR are reviewed individually for the agreed scope.
- 01Data sources
Local files and knowledge stores run on your infrastructure. External sources and mailboxes remain with their respective provider.
- 02Permission check
Roles, ACLs, and organization context determine which information is available.
- 03chifty
The application processes only the context authorized for each request.
- 04Local AI
A self-hosted AI model can perform inference inside your network.
- 05Authorized answer
The answer uses only information the requesting user is allowed to access.
Explicitly enabled AI endpoints
If you choose external models, only requests configured by you are transmitted to that endpoint.
Security by Architecture
Control is part of the architecture.
Access is not decided by a badge. It is enforced technically in every relevant data operation.
Data Sovereignty
The application, data layer, and local AI components reside in infrastructure controlled by you.
Customer-controlled infrastructureTenant Isolation
Data queries are consistently restricted to the active organization context.
Organization-scoped queriesRole-Based Access Control
Roles and capabilities govern administrative and organizational permissions.
RBAC · CapabilitiesFine-Grained Access Control
Access is additionally evaluated by user, owner, project, and permission level.
User · Owner · Project · RolePermission-Aware RAG
Organization and document permissions are applied during knowledge retrieval.
ACL filtering before generationAudit Trail
Administrative and knowledge-related activities are captured in audit and activity logs.
Local audit dataPrivate AI
Your models. Your rules.
Application hosting and AI processing are planned separately. You agree supported models, endpoints and network rules; fully local processing is validated for your tasks and infrastructure.
Local Inference
Supported AI models can run within your own infrastructure.
Inference on your infrastructureModel Sovereignty
You choose the model, endpoint, and network rules that apply to it.
Customer-selected endpointZero Data Egress
Zero Data Egress requires local operation of all necessary models, embeddings, document and supporting services, and the sources in use. External connections are documented separately.
Fully local configurationControlled AI Endpoints
External models are accessed only through endpoints you explicitly configure and approve.
Explicit outbound pathSelf-Hosted RAG Pipeline
Knowledge processing, vector search, and permission filtering can run on your infrastructure.
Local knowledge stackModel Flexibility
AI access remains decoupled from a single model vendor and can be adapted to your environment.
Configurable model accessIdentity & Access
The AI does not automatically see everything.
chifty considers organization boundaries and existing access rights when selecting relevant information.
Organization-bound requests
Every relevant request is associated with an active organization and user context.
Roles and capabilities
Administrative actions require the appropriate role and specifically assigned capability.
Content-level permissions
Projects, lists, documents, and content can carry additional user and owner permissions.
Permissions before the answer
Unauthorized knowledge entries are filtered during search, before the AI generates an answer.
Encryption & Application Security
Concrete measures. Precisely scoped.
We state the actual scope of every measure rather than making blanket security claims about all data.
TLS Encryption in Transit
Browser and application connections are protected by TLS during transmission.
Encrypted in transitAES-256-GCM
IMAP passwords as well as booking and appointment tokens are stored with authenticated encryption.
Scoped credential encryptionArgon2id Password Hashing
Local passwords are hashed with Argon2id and are not stored in plain text.
Memory-hard password hashingParameterized Queries
Database access uses parameterized queries rather than client-composed SQL input.
No client-built SQLSecrets Redaction
Tokens, cookies, and authorization headers are filtered from application logs.
Sensitive log controlsOffline License Verification
On-premise activation is verified locally and requires no ongoing connection to chifty.
No license server requiredOperational Control
Your infrastructure. Your decisions.
On-premise does more than reduce data flows. It gives you control over operations and dependencies.
No Phone-Home
After setup, there is no ongoing data exchange with chifty servers.
Customer-Controlled Infrastructure
The network, storage, database, and runtime environment remain under your control.
Customer-Controlled Updates
You decide when supplied updates are installed within your operating processes.
Reduced Third-Party Exposure
A fully local configuration reduces external providers in the operational data path. Connected cloud sources, mailboxes and explicitly approved external services still need separate consideration.
Infrastructure Independence
Your data and installation remain in your environment, even without ongoing vendor availability.
Private AI by design
Private AI starts with control.
Plan Private AI for a smaller business too. Together we agree supported local models, sources, hardware and operational responsibilities. chifty is in development; scope and availability are agreed for your project.
Frequently asked questions about on-premise security
What stays inside our company with chifty On-Premise?
The chifty application, data layer, and self-hosted knowledge components run on your infrastructure. If you also configure local AI models and local supporting services, operational AI processing can remain entirely within your network.
What does Zero Data Egress mean on this page?
Zero Data Egress applies to a fully local configuration of the agreed scope: chifty, knowledge processing, embeddings, OCR where needed, supporting services, AI models and the sources in use run within your network. External AI endpoints, cloud mailboxes and other external sources have their own data flows and need separate review.
Can we still use external AI models?
Yes. You decide which AI endpoints may be reached. External models are a controlled option, not a requirement for fully local operation.
How does chifty prevent unauthorized AI answers?
Organization, user, and document permissions are applied during knowledge retrieval. Unauthorized entries are filtered before generation.
Does the on-premise installation require an ongoing connection to chifty?
No. Activation is verified locally. After setup, there is no ongoing data exchange with our servers.
Can chifty operate entirely without internet access?
Yes, when chifty is configured with self-hosted AI and all required local services. Some optional features may require external services and are therefore reviewed together before rollout.