Docs

Terms

Definitions for AI agents, context, memory, profile atoms, provenance, evaluation, taste and governed updates.

Agent#

A system that uses a model, tools, memory and control logic to pursue a task or goal across multiple steps.

Approval#

A positive human signal indicating that an output, decision or reference should carry future weight. Approval is evidence, not an automatic universal rule.

Boundary#

A constraint inferred from a rejection or correction. Boundaries should be scoped to the context in which they are valid.

Context#

Information available to the system for the current task. Context may be temporary and does not necessarily become memory.

Continual learning#

The broader machine-learning problem of improving from a stream of experience while preserving useful prior knowledge and avoiding destructive interference.

Correction#

A human change to an output or decision. A correction becomes useful learning only after the system interprets what changed and tests whether that interpretation helps later.

Evidence#

The source material supporting an observation or profile claim: a reference, message, edit, decision, outcome or explicit statement.

Evaluation#

A process for determining whether an output or system change improved performance according to defined criteria.

Feedback#

Information about the quality or outcome of a system’s action. Feedback may be explicit, such as a correction, or structured from an observable result.

Guidance pack#

A task-specific compilation of the most relevant profile claims, evidence, constraints and instructions supplied to an agent.

Iterative AI#

An AI system whose future behaviour changes through a governed loop of experience, evidence, guidance, evaluation and update.

Judgment#

The process of choosing between plausible options according to context, standards, consequences and taste.

Memory#

Persisted information that can be written, managed and read across interactions. Memory supports learning but does not guarantee it.

Observation#

A structured statement extracted from evidence before it becomes a durable preference or rule.

Open source AI#

Under the Open Source Initiative definition, an AI system that provides the freedoms to use, study, modify and share the system, together with the preferred form needed to make modifications.

Open-weight model#

A model whose trained parameters are available for download and use. Open weights do not automatically mean the complete AI system is open source.

Profile atom#

A small, scoped, evidence-backed claim about a preference, boundary, standard, goal or working pattern.

Provenance#

The trace connecting a learned claim back to the evidence and process that produced it.

RAG#

Retrieval-augmented generation: retrieving relevant information and placing it into model context to support generation.

Revocation#

Removing or disabling the future influence of a memory, profile atom, source or learned rule.

Taste#

Contextual judgment about what is worth selecting, preserving, combining or rejecting. Taste is dynamic and shaped by exposure, work and changing intent.

Update proposal#

A suggested change to the system’s profile or behaviour that can be inspected and governed before becoming durable.