The Operating System

The Methodology — Contextual Engineering

The LLM is a processor. The product is the software around it. What makes a processor useful is assembling, at the moment of the task, the right four things — and then writing the result back into the same system it read from. That last step is what compounds.

LLM as Processor — Three Access Patterns

These are access patterns, not product tiers. A "real agent" is not better than a one-shot — it is more expensive and less predictable. Choose the cheapest pattern that closes the task.

The MCS Spine — Git and SQL, on Purpose

The Modular Context System is the durable layer the fleet reads from. It exists twice, deliberately, because the two substrates are good at different things. The rule: author in files, retrieve and telemeter in a database.

Skills — What Made Knows How To Do

A skill is a packaged capability: how to do one thing Made's way, written down once and reusable. Every row carries its own classification — capability, domain, maturity, priority, and the harness it prefers — per the front-matter contract.

Tools — the Fourth Piece

A skill says how. A tool is what it acts on: an external surface with an API, a credential, and usually a bill. Until 2026-08-04 Made had no tool registry at all — the surfaces were real and paid for, but nothing enumerated them.

Brand & Content Spine

The brand bible is the immutable layer: tools and skills change constantly, the brand does not. Made runs 4 packs — not one per client, because Made is a single company, but one per audience surface. A content skill is correct only when it loads the right pack through the standard contract rather than hardcoding a colour.

Harnesses & the Fleet

Once you know the context, skill, tools, and shape, the harness is the environment that runs it: process model, state, retries, credentials, observability, cost. Made keeps a menu of nine patterns and picks the smallest one that fits, rather than inventing a shape per agent.

Retrieval & Graph

The 2026 consensus is that retrieval is a three-stage pipeline — retrieve (keyword + dense) → fuse → rerank — and that a graph earns its place only when questions are genuinely multi-hop. Made is further along here than most builds of its age, so this section is less apologetic than the equivalent one elsewhere.

Tech Stack & Infrastructure

For an agent to have a place to live, be hosted, and hold a domain, there has to be a server. Made runs on one: bc-made. Everything below is that answer, plus the database estate behind it.

Process Traces

Worked end-to-end paths through the system — the methodology above, instantiated on Made's actual data. Each trace names the real Edge Function, RPC, or cron at every step, which is what makes them reviewable rather than decorative. Most follow the data; MADE-ID Custodian is the one that follows ownership and rule — who is allowed to decide two records are the same thing, and on what evidence.

Where a trace follows the data, an anatomy follows one agent all the way down — its prompt, its tools, its budget, its failure paths. Three exist, and the third describes an archetype rather than a single agent. They are the most detailed pages on this site, and all three are hand-authored rather than generated, because the interesting parts are in the source, not the registry.

The traces below are authored in the wiki and are the next candidates to become diagram pages. They are listed here because the written version already exists and is accurate — the visual is the missing half, not the understanding.

Gaps — What Is Not Built

An explicit list, mostly computed rather than typed. This section exists because a map that only shows what works is a brochure. The audience is internal, so everything below is real.