Plain explanations of the systems we build
Short, honest explanations of the technology HDS works with, written for people evaluating whether they need it rather than for people already sold on it. Each one covers what the thing actually is, how it works once it is running unattended, and where it is the wrong tool.
3 explainers
More will be added as the questions come up in real conversations, which is where every one of these started.
- 01 · MCP
What is an MCP server?
How AI tools connect to business systems, what the server controls, and why access needs explicit limits.
- 02 · AGENTS
What is an AI agent?
What an agent actually is, how the loop works, and what separates one from a chatbot or a single model call.
- 03 · RETRIEVAL
What is retrieval augmented generation?
How an AI system answers from your own documents, how the pipeline is assembled, and what it has to be designed against.
How these are written
Every explainer here is written from a system HDS actually runs, not assembled from other people's articles. That constraint is the entire reason this section exists rather than a list of links: the mechanisms described are ones that have been built, deployed, and then lived with afterwards, which is where the interesting detail comes from.
HDS runs its own multi-agent platform for delivery, with persistent memory, review and verification gates as first-class agents, and audit-grade logging of every action, and has shipped retrieval over a private corpus with citations back to source threads in a setting where confidentiality governed every design decision. The systems these are drawn from are described on the work page, alongside the rest of what HDS does.
Each one also says where the approach does not fit. A page that only sells is not worth citing and is not much use to somebody genuinely trying to decide, so the limits are stated as plainly as the capabilities are.
Read one, then tell us what you are actually trying to fix.
Start with a free scoping conversation with Mike Hyams, the person who builds and supports the work.