What is Broks Forge?

Broks Forge is an AI Engineering Operating System. It records the engineering act behind an AI system — what was built, what was measured, what was decided, and why — and then reasons over that record to answer engineering questions.

It is not an evaluation tool, an observability dashboard, or an LLM tracing platform. Those tools answer what happened. Broks Forge is built to answer why it happened, what it means, and what to do next.

The one-paragraph version

You register your AI artifacts — agents, prompts, datasets, providers, models. You evaluate them against real data. From that real work, Broks Forge derives an engineering record: observations of what was measured, claims about what is true, decisions that were taken, evidence that supports them, and durable knowledge that emerges. It versions every artifact with the rationale behind each change (AI Git), maps the whole system as a graph (Forge Graph), and puts two reasoning applications on top: Brok, an engineering partner that answers questions from that record, and the Root Cause Explorer, which assembles a complete investigation when something fails.

Who it is for

Why it exists

Teams shipping AI systems accumulate decisions faster than they can record them. A prompt is promoted on a Tuesday because it "seemed better." A dataset is regenerated. A model is swapped for cost. Six weeks later an evaluation fails, and nobody can reconstruct which of those changes mattered — because the reasoning lived in a chat thread, a pull request comment, or somebody's head.

Observability tools do not solve this. They record traces of execution with excellent fidelity, but a trace cannot tell you that a prompt was promoted without evidence, that a claim contradicts the evaluations behind it, or that this exact failure happened nineteen days ago and the team rolled back. Those are engineering facts, and they need an engineering data model.

Broks Forge is that data model, plus the reasoning layer that makes it useful.

What makes it different

It derives, it does not ask you to author. No one writes an "observation" or a "decision" by hand. Promoting a prompt version is a decision; running an evaluation against it is evidence. The reasoning objects are computed from work you were doing anyway.

Nothing is fabricated. The reasoning layer is a deterministic engine over real database rows, not a language model summarising documentation. If the record cannot answer a question, the answer is "the engineering record cannot answer that" — followed by the questions it can answer.

Every statement declares how it is known. Each one is marked derived (read directly from records), inferred (a causal reading that could be wrong), suggested (a recommendation), or unknown. Confidence is a three-step verbal ladder — consistent with, likely, near-certain — never a fabricated percentage.

Absence is never dressed up as health. An artifact nobody has evaluated is reported as unproven, not as passing.

The shape of the system

                    ENGINEERING APPLICATIONS
        Brok (partner)  ·  Root Cause Explorer (investigation)
                              |
                        FORGE GRAPH
              the living map of the whole system
                              |
                          AI GIT
          revisions, promotions, rollbacks, rationale
                              |
                         REGISTRY
        one catalog of artifacts and derived knowledge
                              |
                       FORGE KERNEL
            the invisible foundation everything sits on

Read The Five Layers for what each one owns.

What it is not

• Not a chatbot with a dashboard. Brok has no language model behind it. It resolves an

engineering intent and composes an answer from rows that exist.

• Not a tracing tool. There is an Execution Graph, but it is reconstructed from an evaluation's

own recorded runs to answer "where did the chain break?", not to sample production traffic.

• Not a replacement for your framework. Agents are registered by endpoint. Broks Forge is

framework-agnostic and does not ask you to adopt a SDK to be useful.

• Not a benchmark leaderboard. Evaluations exist to produce evidence for decisions about your

system, not to rank models in the abstract.

Where to go next

• New to the category → The AI Engineering Operating System
• Want to run it → Getting Started
• Want the object model → Core Concepts
• Comparing tools → Comparisons Overview
• Building on it → Architecture Overview