Your data stays with you. No exceptions. By architecture.

No — and not by policy, but by architecture. Models, knowledge layer, and agents run entirely on hardware within your own infrastructure. There is no cloud connection and no call to external APIs: inference, indexing, and storage happen exclusively inside your network. What technically has no path to the outside cannot leak — not documents, not queries, not metadata.

Lokalaise is a managed local AI stack: your company data is connected through a connector layer (file shares, SharePoint, DMS, email, PDF scanning), indexed permission-aware in the knowledge layer, and made usable through enterprise chat, agents, and knowledge workflows. All layers — including the language models themselves — run on dedicated hardware located on your premises. Data at rest is encrypted with AES-256; internal connections are secured with TLS 1.3.

Lokalaise adopts the existing roles and access rights from your source systems and enforces them down to the document level. The check happens before an answer is generated: documents the person asking has no access rights to never enter answer generation in the first place. That way, no summary, preview, or indirect phrasing can disclose content — salary data to roles outside HR, for example. When permissions change in the source systems, the new state applies in the knowledge layer as well.

Every interaction leaves a complete audit trail: who asked which question and when, which sources and document versions were used, and which permission check took place. In addition, every answer visibly carries its evidence — document, page, version state. Like everything else, the logs live in your infrastructure and are exportable, so your IT, your data protection officer, or external auditors can analyze them independently.

Operations and maintenance are part of the managed stack: we deliver vetted updates for platform and models as controlled releases that are applied deliberately — not as a silent cloud rollout. You stay in control of when an update enters your environment. In isolated networks, updates are applied via offline transfer; a permanent internet connection is never required for operation.

Yes. The entire stack is air-gap capable and can be operated in fully isolated networks without any internet connection. Inference, indexing, permission checks, and audit logging require no external connection — the models run locally and access only your internal knowledge. This matters for environments with particularly sensitive data, where even outbound traffic must be ruled out.

Processing happens exclusively on-premise, within your sphere of responsibility: there is no third-country transfer, no sharing with external model providers, and no use of your data for third-party training. Encryption (AES-256 at rest, TLS 1.3 in transit), document-level permissions, and exportable audit trails provide technical and organizational measures you can concretely demonstrate to data protection officers and auditors.