EU-Based · EU-Hosted · Zero Data Residency Risk
Self-hosted AI for teams that can't
send their data to OpenAI.
A production AI stack on your infrastructure. Compliance-defensible, cost-predictable, fully documented. Sovereign by default — no cloud dependency, no vendor lock-in, no mandatory retainer.
31 containers · 6+ LLM providers · 4 OCR engines · 13 tool integrations · Operator-grade DR runbooks
A pilot self-hosted AI deployment that your Legal, Security, and Finance teams can actually defend. From €5K–€10K for a foundation engagement — because we don't have a sales org, a San Francisco office, or a €30B valuation to justify.
Who this is for
If your data can't legally leave the building, you're our buyer. Common engagements:
Healthcare & MedTech
Patient data, clinical notes, imaging metadata that can't transit US providers under GDPR Art. 9 or sector-specific rules. Self-hosted LLM on EU hardware with audit trail.
Legal & Professional Services
Privileged client documents, M&A diligence, regulatory filings. Attorney-client privilege survives intact because the data never leaves your perimeter.
Financial Services & Fintech
NIS2, DORA, EBA outsourcing guidelines. Auditor-ready evidence pack, cost-predictable for CFO sign-off, no per-token cloud surprises mid-quarter.
Not on this list? Defence contractors, pharma, public-sector teams, and academic groups face the same data-residency and compliance constraints. Tell us what you're building →
Three traps. Three exits.
Vendor lock-in
Your AI stack is married to a single cloud provider. Migration cost grows with usage. Sovereignty review fails.
→ Swap any layer. Keep everything else.
6+ LLM providers behind one OpenAI-compatible API. Provider swap is a config diff — application code never changes. Your sovereignty review passes.
Model churn
Your provider deprecates a model mid-engagement. Behavior shifts silently. Compliance evidence rots.
→ Provider swap in minutes, not months.
Route to a different provider with a config change. When byte-reproducibility is required, local models with frozen weights are available. No surprise upstream changes reach your production stack without your approval.
Governance opacity
"Our AI is in the cloud somewhere." No control plane. No reproducibility. No defensible answer for the regulator.
→ Single config tree. Every change is a commit.
Git-versioned configuration. Pre-flight checks refuse to render on drift. The regulator gets a defensible answer.
These aren't theoretical. They're why your compliance team is blocking AI adoption right now.
What's actually running
All 49 components are independently restartable · systemd-supervised where stateful · provider calls fronted by an OpenAI-compatible router with per-call cost tracking · sub-2h bare-metal recovery, tested.
Documentation you can audit
Every CPLT engagement ships with operator-grade documentation. Not a dashboard and a support email — actual runbooks, decision records, and recovery procedures your team can verify, extend, or operate independently.
DR Playbook — Public Summary
Public overview of our disaster recovery standard: recovery time objectives, backup architecture, decision tree methodology, and verification-first philosophy. Shows the documentation quality every CPLT engagement receives.
Read the DR summary →Full DR Playbook (54 pages)
Detailed recovery procedures, verification commands, and operational runbooks. Shared under NDA during scoping conversations — part of every engagement deliverable.
Scope a deployment →What you get
Standard package: LibreChat stack + document processing pipeline + operator-grade DR documentation. Scoped and priced in a written proposal.
Add-ons: Sovereign OCR, custom MCP servers, tailored integrations, observability stack, off-site backup pipeline — scoped per engagement.
No mandatory retainer. No time-and-materials surprises.
Watch it deploy
A simulated deployment — same flow used to provision the production stack on CPLT’s own infrastructure.
Ready to build AI you actually own?
Tell us what you're trying to build. We'll respond within 5 business days with a written proposal — or an honest "no" if your need doesn't match what we do.
Better to be told that early than to be retainer-walled into the wrong relationship.