Skip to content
XZMADE SYSTEMS
Menu

Services · AI & agents

Practical AI, wired into how your business works

We connect AI to your documents, systems and tools, and run it on infrastructure you control. It answers from your own material, acts only through tools you approve, and keeps a record of what it did.

What AI can and can't do

What AI does well

  • Finding and summarizing answers across large amounts of text, with links back to the source.
  • Drafting replies, reports and code for a person to check.
  • Carrying out routine, multi-step tasks through well-defined tools.
  • Turning a plain-language request into a structured action.

Where it needs a person

  • It can be confidently wrong, so answers should cite sources and important actions need review.
  • It only knows what it's given. Missing or outdated documents mean missing or outdated answers.
  • It shouldn't make final calls with legal, financial or safety consequences.
  • Messy, unmaintained data usually needs cleanup first.

What we build

01

Internal AI assistants

Company-wide assistants connected to your source code, documents, knowledge bases and project tools. Staff ask questions in plain language and get answers grounded in your own material, with links to where each answer came from.

  • Connected to the places your knowledge actually lives, such as file shares, wikis, code repositories and ticketing tools
  • Access follows your existing permissions, so people only get answers from what they could already open
  • Answers cite their sources, so they can be checked

02

Customer-facing chatbots

Assistants that answer customer questions using your product data and documentation, available when your team isn't.

  • Clear boundaries on what the assistant can see and do
  • Handoff to a person when a question is out of scope or the customer asks for one
  • Conversations logged so you can review answers and improve the source material

03

AI agents & automation

Agents that take actions, not just chat. They work through tools and APIs you define, and changes are staged for review before they go live.

  • Agents for routine back-office work across the systems you already use
  • Coding agents and agent harnesses for development teams
  • Staged changes with human review, health checks and rollback

We run this approach in production in our own platform: Spellship

04

MCP servers & integrations

Custom Model Context Protocol (MCP) servers that connect AI assistants such as Claude, ChatGPT and others to your business systems and SaaS tools, with the same permissions and logging as any other integration.

  • One integration that works across the AI assistants your staff already use
  • Read-only or read-write access, scoped per tool
  • We've published open-source MCP work of our own

05

LLM gateways & model operations

One API in front of multiple AI providers and your self-hosted or local models, so applications aren't tied to a single vendor.

  • Routing per task, with automatic failover when a provider is down or slow
  • Caching and rate-limit handling
  • Usage tracking and cost limits per team or application

06

Self-hosted & private AI

Open-weight models running on your own hardware or on Canadian infrastructure, for data that can't leave your control.

  • Hardware sizing and model selection for the job, not the biggest model available
  • Credentials kept in a vault and used on the agent's behalf, never placed where the model can read them
  • An honest comparison of quality, speed and cost against hosted models

07

AI simulations & training tools

Scenario simulators driven by language models, combined with real map data, for training and exercises.

  • Scenarios that respond to what participants decide, instead of a fixed script
  • Real geography from map data, so exercises reflect actual places
  • Facilitator controls and a record of each run for debriefing

08

AI readiness & advice

Before building anything, figuring out where AI actually helps your organization, and where it doesn't.

  • Use-case review against how work is done today
  • Choosing models and providers for each job
  • Data, privacy and security considerations
  • A rollout plan with guardrails and a way to measure whether it's working

How we deploy AI responsibly

  • Secrets never reach the model. Passwords, keys and tokens are kept in a vault and used on the AI’s behalf, never shown to it.
  • People approve what goes live. Changes are staged and reviewed before they reach production.
  • Everything is logged. You can see what the AI was asked, what it answered and what it did.
  • Your data stays where you need it. Hosted, self-hosted or Canadian infrastructure, chosen for your requirements.
  • We’re honest about limits. If AI isn’t the right tool for a job, we’ll say so.

Questions we hear often

Will our data be used to train models?

Not by us. We use providers' business and API offerings, which generally don't train on customer data by default, and we check each provider's terms and settings before using it. When data can't leave your control at all, we run open-weight models on your own servers or on Canadian infrastructure.

Can it run on our own servers?

Yes. Open-weight models can run on your hardware or in a Canadian data centre. We'll size the hardware and tell you plainly how the results compare with hosted models for your use case.

Which AI models do you use?

We're model-agnostic and pick per job. A large hosted model may suit complex reasoning, while a small self-hosted model may be faster, cheaper and private enough for routine work. A gateway lets you switch later without rewriting your applications.

What happens when the AI gets something wrong?

It will sometimes, so we design for it. Answers cite sources, actions are staged for review, everything is logged, and changes can be rolled back.

Do we need a big project to start?

No. Most organizations start with one assistant or one workflow, measure it against how things work today, and expand from there.

Who owns what you build?

You do. The code, configuration, prompts and data live in your repositories and accounts.

Tell us about your project

Tell us what you'd like AI to do, and what data it would need. We'll tell you honestly whether it's a good fit.