Claude Enterprise
Anthropic's enterprise platform — Claude Code-style agents, strongest reasoning for complex multi-step work, enterprise data controls. My default recommendation when the workload justifies it.
AI Integration · your accounting, CRM & email
Put AI to work on the jobs
Put AI to work on the jobs
nobody likes doing.
Not another chatbot in a browser tab. I set up an AI assistant that plugs into the software you already use — your accounting or ERP, your customer records, your Outlook — and does the slow, repetitive work: looking things up, drafting invoices and emails, preparing reports. It always asks before it sends anything, and it can run entirely on hardware you own.
Claude · OpenAI · self-hosted Approval-gated writes Every action logged
The core offering
An agent on your ERP server, doing the busywork
Most business AI stops at a chat window. This doesn't. I put an agent on the Windows server where your ERP lives — with least-privilege access to the database and the file share — so it can do real work: query orders, reconcile deliveries, generate invoices and reports as files, draft the follow-up email in Outlook. Think of a capable engineer at that terminal, except it's tireless, logged, and never acts without sign-off.
- Works with the ERP you already run — Sage, Microsoft Dynamics, NetSuite, SAP Business One, QuickBooks, or a custom in-house system. Old and on-prem is often the easiest — if it has a database or an API, it can be integrated.
- CRM & Outlook too — Dynamics, HubSpot: customer histories summarized before a call, quotes drafted from CRM data, inboxes triaged with your rules.
- Automations that run in the background — scheduled jobs the agent handles end-to-end, surfacing only what needs a human decision.
Where it plugs in
Integration points I build against, in the order clients ask for them.
- ERP
- Sage, Dynamics, NetSuite, SAP Business One, QuickBooks, or custom. Direct database access on the on-prem ones, API on the cloud ones — least-privilege either way: order and inventory queries, invoice generation, reconciliation, reporting, with writes behind approval gates.
- CRM
- Dynamics, HubSpot, SQL CRMs. Customer summaries, quote drafting, pipeline hygiene, data cleanup — against API-based and database-backed systems alike.
- Outlook / M365
- Drafting, triage, and follow-ups grounded in your ERP and CRM data — the reply cites the actual order status, not a guess.
- File servers
- The agent reads and writes real files on your share — Excel reports, PDFs, import/export batches — where your team already looks for them.
- Anything with an API
- Ticketing, accounting, scheduling, industry-specific SaaS. If it exposes data, it can join the workflow.
No lock-in. No black box.
The integration layer is yours; the model behind it is swappable. Three ways to run it:
OpenAI Enterprise
GPT-class models with enterprise agreements. If your organization is already committed here, the same connectors and guardrails apply — the integration doesn't care.
Self-hosted models
DeepSeek, Llama, Qwen and friends on a GPU server you own. Prompts and data never leave your building — the right answer for regulated data, and often cheaper than per-seat licensing.
Trust, engineered
Agents with database access need adult supervision
Giving software the keys to your ERP is a security decision, and I treat it like one — the same discipline as the EDR platform I run.
- Read-only first. Every integration starts with SELECT. Write access arrives workflow-by-workflow, after you've watched it be right.
- Approval gates. Outbound email, database writes, and file changes queue for a human click — until you decide a workflow has earned autonomy.
- Least privilege. The agent gets its own database account scoped to exactly the tables it needs. Not sa, not domain admin.
- Full audit trail. Every query, file, and draft is logged where you can read it.
- Access
- Dedicated scoped DB account · no shared credentials
- Writes
- Approval-gated · reversible · staged before commit
- Logging
- Every action recorded · tamper-evident · yours
- Data
- On-prem model option — nothing leaves the LAN
Also on offer
Private AI hosting: chatbots & agents on your hardware
For teams that want AI fully in-house, I build and run private LLM servers — small, fast models tuned to reply in real time, plus the reverse proxy, auth, and monitoring around them.
Company chatbot
A private assistant grounded in your documents — policies, wikis, SharePoint, ticket history. Small open-weight models keep replies quick; your data keeps its residency.
Task agents
Background workers for repetitive jobs — document intake, summaries, data extraction, report prep — running entirely on your own GPU server.
The full stack, managed
Hardware sizing, model selection, hardened hosting, audit logging, and upgrades. One-shot build with handover, or kept running by me afterwards.
Where this is headed. Every private deployment I build runs on hardware in Canada. Longer term, I am working toward shared Canadian-owned GPU infrastructure — so a BC business can rent serious private-AI capacity without sending its data across the border or buying a server outright. That's a direction, not a product page yet; the on-prem builds above are what I ship today.
Proof of work
I run this stack myself, every day
The tooling I sell is the tooling I operate. The ERP Agent Console is a live demo of this exact offering — an agent with least-privilege access to a sample ERP database, approval gates and all, running on my own GPU server. LLM Chat is a private assistant on the same hardware — nothing leaves the LAN. Tower (shown here) pairs a live SSH terminal with an AI coding assistant; it's how I manage this infrastructure from anywhere.
The ERP demo is password-protected — ask for access and I'll send you the credentials. If you'd rather see the idea first without signing in, there's a short scripted walkthrough of the assistant.
How an integration lands
Fixed scope, small first step, expand on proof.
- Check-up & pick one workflow Free 60-minute review of your ERP/CRM landscape. Together we pick the single workflow with the best payoff — usually something everyone hates doing manually.
- Pilot in weeks, read-only A pilot integration, scoped and quoted in writing up front, that you can watch working against real data before it's allowed to change anything.
- Expand workflow by workflow Writes get enabled where trust is earned; new workflows get added where the pilot proves value. You own everything that's built.
Questions I get asked
Our ERP is 20 years old and the vendor barely supports it. Can you still integrate AI?
Usually, yes — old systems are often the easiest, because everything lives in a SQL database I can read directly. No modern API required.
Which AI vendor should we use?
Whichever fits your data and budget: Claude Enterprise or OpenAI Enterprise when you want frontier capability with enterprise controls, self-hosted models (DeepSeek, Llama, Qwen) when data can't leave, or a mix. The integration is built so switching later is a config change, not a rebuild.
What stops the agent from wrecking our database?
Scoped credentials (it can only see the tables it needs), read-only defaults, approval gates on every write, and a full audit log. Pilots run read-only until you've verified the output yourself.
Does our data get used to train someone's model?
No. Enterprise agreements from Anthropic and OpenAI exclude training on your data — and with self-hosted models the question doesn't even arise, because nothing leaves your network.
What does a pilot cost?
Scoped and quoted in writing after the free IT check-up — the quote depends on the system and the workflow. The check-up itself costs nothing and you keep the written report either way.
Talk about an AI integration
Tell me what ERP or CRM you run and which chore you'd automate first. I'll reply within a business day with whether it's integrable, a rough shape, and a cost.
- Your website, domain and email
- Your computers and updates
- Your backups
- How people sign in
- A quick look at your network
Prefer email? Write to info@sd-techsolutions.com. I reply within one business day.