Cloud Exit Blueprint
Not "install Ollama". A migration plan.
If you've built something real on LLM APIs — a support bot, a content pipeline, a summariser — you already know the three pains: the bill scales with usage, a rate limit caps a demo, pricing changes without your vote. This blueprint is the structured plan a small agency or technical team uses to move off the API without breaking the product.
What's inside
- 📄 The Blueprint (24-page PDF)
- Reference architecture: API-native -> hybrid (policy gateway) -> fully local, drawn as real diagrams
- Vendor lock-in risk checklist (3 pages): data portability, prompt portability, rate-limit + ToS + pricing risk
- Migration sequencing template: batch/offline before real-time, non-critical before critical
- Decision framework: frontier reasoning and newest-model tasks stay cloud, by design
- 3 worked scenarios (RAG support bot, content pipeline, batch report summariser) with step-by-step targets
- Honest 90-day plan + troubleshooting + cost-framing math you fill with your own numbers
- 📊 The Workbook (XLSX, 2 fillable tabs) — workload inventory + risk-scored sequencing
- 🖨️ Cheatsheet (2-page PDF) — the decision framework as a 1-page flowchart
What this is NOT
It is not the $0 AI Stack Playbook. That entry-level product (ENTRY tier, $15) replaces an individual freelancer's cloud subscriptions with local models — the "who should I replace, how much do I save" kitchen. This blueprint is the technical follow-on: a team that has already built something real on APIs, and wants to architect the exit with a risk checklist and migration sequencing. Different buyer, different depth — buy the playbook first if you're the one-person version.
Who this is for
Agencies and technical teams (2-20 people) with a real product or client workflow on OpenAI/Anthropic/Google APIs who are worried about cost scaling, rate limits, or vendor lock-in as usage grows. You can run Docker, you have an eval mindset, and you are OK to keep cloud where it genuinely wins.
Who it's NOT for
Solo freelancers just cutting personal subscriptions (that's the $0 AI Stack Playbook) or teams that want a one-click migration rather than a planned one. If your workloads are all frontier reasoning with no batch, the honest answer is "stay cloud" — this blueprint will tell you that too.
Why trust this?
No fabricated savings statistics: every workbook cost cell is filled from your provider console, and the guide refuses to print "you'll save 70%". The worked scenarios are the shapes we see in real client work — generic, specific, and no invented case studies. Build it to the guide at local, and review the checklist quarterly.