ENGAGEMENT MODEL

Three stages. Fixed scope. Zero ambiguity.

Every engagement follows the same rhythm: define the scope, build and verify, hand over and clean up. No surprises. No scope creep. No lingering dependencies.

The three stages

STAGE 01

Scope & Sandbox

1—2 weeks
  • Written proposal with fixed scope and price
  • Sandbox environment provisioned on your hardware
  • Strict isolation — no access to production systems
  • Clear acceptance criteria defined before any build work
Deliverable
Signed scope document with: exact capabilities (core stack + add-ons), acceptance criteria per capability, fixed price (not time-and-materials), add-on pricing broken out separately, explicit out-of-scope items.
No Stage 2 begins until scope is signed
STAGE 02

Build & Verify

Per scope (typically 2—8 weeks)
  • Capabilities built per signed scope
  • Deterministic mapping: every config → every container → every unit
  • SSOT compiler ensures reproducibility
  • Verification against acceptance criteria
Deliverable
Working capabilities per scope, verification report (what was built, how it was verified), and the configuration tree (git-versioned).
No Stage 3 begins until all acceptance criteria pass
STAGE 03

Handover & Shred

1 week
  • Full handover: source code, IaC, runbooks, verification tools
  • Knowledge transfer sessions (recorded)
  • All CPLT access to your systems revoked
  • All CPLT-held data from your engagement shredded
Deliverable
Complete source repository, Infrastructure-as-Code (Docker Compose, systemd units, scripts), operator runbook (your team maintains independently), and a shred certificate confirming all CPLT-held data is gone.

After Stage 3, CPLT holds nothing from your engagement. No code, no configs, no logs, no backups. Your sovereignty is complete.

Choose your engagement shape

Most common

Programmatic

1—3 months — multi-capability

Examples Full stack standup with MCP tool fleet and observability layer. Production AI platform with document processing pipeline. Sovereign AI deployment with compliance-grade audit trail.

Best for: teams building from scratch or doing a major upgrade across multiple capabilities.

Entry point

Tactical

≤ 2 weeks — single capability

Examples Add deterministic OCR to an existing document flow. Wire a sovereign LLM behind an internal application. Build a cost dashboard for a multi-provider setup.

Best for: teams with an existing stack that needs one specific capability added.

Optional

Continuous

Monthly — no lock-in

Examples Ongoing optimization of a production AI stack. Quarterly compliance re-runs with delta-only reassessment. One-operator extension of your team without hiring overhead.

Best for: buyers who want long-running operator support without the coordination tax of a larger team.

Continuous engagements are optional — no lock-in, no auto-renewal. Each month is a new fixed-scope agreement. cplt.online supports all three shapes; cplt.tech and cplt.studio do not offer Continuous.

What an engagement looks like

Three sample engagements illustrating the typical scope, timeline, and price range for self-hosted AI deployments in regulated industries.

LEGALTECH 80 people · Berlin

Doc-Q&A over a 12-year case archive

The need: Litigation team wanted ChatGPT-style search across two decades of pleadings, exhibits, and judgments. US providers were off the table — client data, GDPR Art. 28, German bar association rules.

Shape
Programmatic (one-shot)
Stack
LibreChat + Qwen2.5-32B + Qdrant + OCR pipeline + LiteLLM
Hardware
Existing rack (1× RTX 6000 Ada, 48GB) — already on-prem
Timeline
6 weeks (Stage 1 → 3)
Price band
€18K–€25K fixed
Add-ons
Advanced OCR (scanned exhibits, hand-written annotations), DR runbook handoff

Outcome: 40k documents indexed, sub-2s retrieval, full audit log of every query. Lawyer can cite which document the answer came from. Continuous: optional, declined.

Sample engagement scope — illustrative, not a delivered case.

HEALTHTECH 35 people · Lyon

Patient-record summarisation, on-prem only

The need: Clinical team wanted automated discharge-summary drafting from consultation notes. Patient data cannot leave the hospital network — HDS hosting requirement, MDR Class IIa device implications.

Shape
Programmatic + Continuous (3 months post-launch)
Stack
LibreChat + Llama-3.3-70B (4-bit) + structured-output schema + audit logger
Hardware
2× H100 supplied by client; physical air-gap, no internet egress
Timeline
8 weeks build + 3 months monitored stabilisation
Price band
€32K build + €1.8K/month Continuous (8h/mo)
Add-ons
Custom JSON schema validators, observability stack, off-site encrypted backups

Outcome: Every model output is structured, traceable, and auditable. Clinician edits ~30% of drafts vs writing from scratch. Zero data egress.

Sample engagement scope — illustrative, not a delivered case.

FINTECH 120 people · Amsterdam

Internal copilot for a regulated trading desk

The need: Compliance and ops wanted internal chat + code-assist on a shared LLM, but DORA + MiFID II prohibited sending order-flow context to a SaaS provider. Existing GitHub Copilot use was already flagged in audit.

Shape
Programmatic, build-only handoff
Stack
LibreChat + Qwen2.5-Coder-32B + DeepSeek-V3 (via LiteLLM fallback) + Continue.dev integration
Hardware
Client-spec'd: 4× L40S in their colo, dual-region replication
Timeline
10 weeks (includes IAM integration with their existing Okta + audit pipeline into Splunk)
Price band
€45K–€60K fixed
Add-ons
SSO integration, custom MCP server for internal data sources, DORA-aligned DR documentation

Outcome: 95 active users, fully documented for the next DORA audit, zero retainer required. Internal team owns operations from day one.

Sample engagement scope — illustrative, not a delivered case.

Your engagement won't match any of these exactly — that's why Stage 1 exists. The scope document we produce is yours regardless of whether you proceed.

Ready to scope?

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.

Scope your deployment → Read the architecture → Compare options →