Glossary of BX-System Terms

The glossary defines every term used in BX-System documentation and articles.

BX-System

An agent kernel that generates its own tools, verifies them against expected outcomes, learns from the results, and follows five operating principles.

Agent kernel

A runtime that coordinates tools through declarative flows, verifies their behavior, and updates its own strategy weights based on verified outcomes.

Operating principle

One of the five rules BX-System follows: continuous operation, self-protection, adaptation, automatic repair, efficiency.

Cycle

One complete loop of BX-System: generate a tool or flow, verify the outcome, record the result, update the strategy weights.

Context

The body of accumulated knowledge that each cycle adds to. Stored in workflow context and memory state.

AST engine

The component that corrects code that violates a defined rule by parsing it, applying policies, and writing the corrected version.

Stable state

The condition to which the kernel returns after a repair or after a failed verification.

Reward

A scalar value between 0 and 1 computed from actual outcomes, used to update strategy weights in the cognitive memory engine.

Role

A named behavior profile that determines prompt prefix, bias profile, and task routing for a language model call.

Verification

The process of executing a tool against expected outcomes, comparing the actual result with the expected result, and recording the verdict.

Declarative flow

A flow that names the tools and their order. The kernel executes the flow; the flow does not execute itself.

Isolated context

The execution environment where tools run. It uses immutable snapshots to prevent state corruption.