Self-hosted LLM memory for enterprise — GDPR, on-prem, air-gapped¶
TL;DR¶
Two deployment models — self-hosted on your infrastructure, or Versino-managed hosting as a dedicated single-tenant deployment. No multi-tenant public SaaS. Pro and Enterprise tiers unlock additional throughput, the compliance pack, signed deletion certificates, and contracted support.
Hosting models¶
| Model | Operated by | Fit |
|---|---|---|
| Self-hosted | Customer | Maximum control; data never leaves customer-owned infrastructure; air-gapped deployments supported. |
| Versino-managed hosting | Versino PsiOmega GmbH | Dedicated single-tenant deployment operated by us on EU-jurisdiction infrastructure. Patches, monitoring, backups, and SLA included. Useful when you want the engine without operating it yourself. |
Both models run the same wheel and expose the same surface. Pick the model first, then a deployment topology below.
Deployment topologies¶
| Topology | Fit | Notes |
|---|---|---|
| Embedded library | Single Python service, in-process state | Lowest latency. State lives in process memory; persist via to_bytes(). |
| REST sidecar | Polyglot microservices, Kubernetes | semvec serve runs as a sidecar; HTTP from anywhere. Single binary. |
| Multi-region (Cortex) | Distributed agents, regional residency | Cortex REST endpoints (/v1/cluster/*, /v1/region/*) for region pinning and observer-mediated transfers. |
| Air-gapped | High-assurance environments, regulated industries | Self-hosted only. Semvec has no outbound calls at runtime; license validation is offline JWT. ONNX int8 embedder option for on-prem inference. |
Licensing tiers¶
| Tier | For | Headline limits |
|---|---|---|
| Community | Evaluation, research, hobby projects, single developer | Rate-limited; no compliance pack; no SLA |
| Pro | Single product team / single deployment | Per-seat, higher throughput, full feature set |
| Enterprise | Multi-tenant, multi-deployment, regulated workloads | Per-deployment, indemnification, SLA, dedicated support |
For current quotas and pricing, see Licensing, visit https://www.semvec.io, or contact vertrieb@versino.de for Enterprise terms.
Data protection & compliance¶
- Hosting model: Self-hosted on customer infrastructure, or Versino-managed dedicated hosting. No multi-tenant public SaaS exists.
- GDPR Article 17: Built-in deletion flow via the compliance pack —
POST /v1/compliance/users/{user_id}/forgetproduces a signed deletion certificate the data subject can keep as proof of erasure. - Audit trail: Append-only event store
(
semvec.compliance.event_store) with HMAC-chained entries; RS256 signatures for tamper-evidence on shipment. - Encryption at rest: Disk and DB encryption are your platform's
responsibility. Semvec additionally ships
semvec.compliance.encryption(AES-GCM per-embedding with operator-derived keys) for an extra layer at the embedding level. Snapshots carry integrity checksums that detect tampering even without encryption-at-rest. - PII handling: Memory contents can be redacted at snapshot time
(
include_memory_text=False,include_literal_cache_text=False) so you can share state with third-party support without leaking user prose. - Cloud Act exposure: Self-hosted deployments inherit your hosting provider's jurisdiction — choose EU-jurisdiction infrastructure if Cloud Act risk is in scope. Versino-managed hosting runs on EU-jurisdiction infrastructure by default.
Support & SLA (Enterprise)¶
- Named escalation contact at Versino PsiOmega GmbH.
- Response-time SLAs negotiated per contract.
- Indemnification: Enterprise contracts include indemnification
against third-party IP claims directed at the Semvec engine
itself (scope: claims arising from features delivered through
the Python
semvecpackage and the REST API; cap and exclusions negotiated per contract). Scope reflects the patent-application stage (see IP & patents below) and is reviewed with each renewal. - Security disclosure intake: security@versino.de.
IP & patents¶
Patent applications pending for the Semvec engine algorithms and functional mechanisms:
| Filing | Number | Status |
|---|---|---|
| European Patent Office | EP 25 188 105 | Filed 2025; pending; within the 18-month confidentiality period under Art. 93 EPC, file not yet publicly available |
| European Patent Office | EP 26 160 795 | Filed; pending |
| United States Patent and Trademark Office | US 19/269,195 | Filed; pending |
Until grant, references to "patent-protected" features describe claims of pending applications, not enforceable exclusive rights.
Public documentation describes the interface and observable behaviour
only. For confidential due diligence on the mechanisms covered by the
pending applications (e.g. procurement security review), arrange an
NDA-bounded review via support@versino.de.
Procurement checklist¶
- Tier picked: Community / Pro / Enterprise
- Deployment topology decided (embedded / REST sidecar / multi-region / air-gapped)
- Embedder chosen (on-prem ONNX int8 / SentenceTransformers / hosted)
- Compliance pack required? (audit, retention, deletion certs)
- Region pinning required? (Cortex)
- Audit / pen-test scope agreed
- SLA terms agreed (Enterprise)
- Patent / IP terms reviewed by legal
Contacts¶
- Sales / Enterprise: vertrieb@versino.de
- Technical support (Pro / Enterprise): support@versino.de
- Security disclosures: security@versino.de
- Publisher: Versino PsiOmega GmbH
- Homepage & pricing: https://www.semvec.io