Sovereign Self-Hosted Local LLM for NetSuite: $0.00 Per Token, 100% On-Premise / Private-VPC Data Residency
The basis-point lever
IT and finance OPEX compression: eliminates $4,500–$9,500/mo in modeled cloud API token and per-page OCR SaaS fees while satisfying strict CISO/audit data-residency mandates.
What you get
- Dual-Tier Docker Compose sovereign inference stack (docker-compose.yml) with internal: true air-gapped bridge network (do_sovereign_inference_net)
- Tier 1 Enterprise Rack / Private VPC profile: vLLM serving Qwen2.5-VL-7B/32B-Instruct (Vision/OCR) and Llama-3.3-70B / Qwen2.5-Coder-32B (SuiteQL & 3-Way Match)
- Tier 2 Single-Node Edge Appliance profile: Ollama + Docling configured for local RTX 4090 / A6000 workstations or Mac Studio M2/M3 Ultra appliances
- Local OpenAI-compatible routing adapter wired into NetSuite MCP (execute_suiteql) and Containerized OCR Authority (do-ocr-authority-worker)
- DefendAir 4-layer OCR prompt-injection sanitizer and AST SuiteQL read-only firewall
Self-Hosted Local LLM vs. Metered Cloud LLM APIs
| Dimension | Metered Cloud LLM APIs & SaaS Wrappers | DataOngoing Self-Hosted Local LLM |
|---|---|---|
| Marginal cost per token / page | Recurring per-1M-token + per-page SaaS markups | $0.00 / token and $0.00 / page on your own hardware |
| 36-month TCO (25k docs/mo + 40 users) | $162,000–$342,000 over 36 months (modeled) | $26,000–$47,300 total ($18,500 setup + compute, 78%–92% savings, modeled) |
| ERP & GL data residency | Prompts and GL rows traverse external third-party APIs | 100% inside your Private VPC or on-premise rack (internal: true) |
| Model weight ownership | Opaque provider checkpoints subject to deprecation | Pinned open-weight checkpoints (Qwen2.5-VL, Llama-3.3, Qwen2.5-Coder) |
Proof behind the offer
- Interactive TCO Calculator: Self-Hosted Local LLM vs. Cloud Token Tax (Tab 5)
- Briefing: Self-Hosted Local LLM for NetSuite — Zero Token Tax & 100% Data Sovereignty
- Compare: Self-Hosted Local LLM vs. Cloud API Token Tax
- Governed AI workers inside NetSuite, with a human approval gate on every write
Frequently asked questions
Does any NetSuite financial, customer, or vendor data ever leave our network?
No. The vLLM and Ollama inference containers bind exclusively to an internal Docker bridge network (internal: true) with telemetry disabled (VLLM_NO_USAGE_STATS=1). Zero bytes of your General Ledger, customer pricing, or vendor banking data ever leave your VPC or on-premise server.
What hardware do we need to run the Self-Hosted Local LLM stack?
We support two production tiers: Tier 1 (Enterprise Rack / Private VPC) runs vLLM on 1x–2x NVIDIA L40S (48GB) or A100/H100 GPUs for 25,000–250,000+ OCR pages/month and 100 concurrent ERP users. Tier 2 (Single-Node Edge Appliance) runs Ollama + Docling on a single RTX 4090 (24GB) workstation or Apple Mac Studio M2/M3 Ultra (128GB Unified Memory) for 2,000–25,000 pages/month.
Can our self-hosted Local LLM power both document OCR and conversational SuiteQL reporting?
Yes. Both vLLM and Ollama expose a drop-in OpenAI-compatible /v1/chat/completions endpoint on your internal network, powering both do-ocr-authority-worker (for Sales Orders, AP Bills, 3-Way Match, and Custom Records) and governed HelpAirs SuiteQL assistants.
Talk to the architect, not a salesperson
HelpAirs technology for NetSuite: the five roles a company pays for, delivered as one supervised AI service at a modeled 20-80% lower cost, with the existing team repurposed rather than replaced, and every outcome published in the results ledger with its source class.
Scope a Self-Hosted Local LLM Sprint Read the results ledger
The direct email and phone number are shared in the receipt of a verified request.