Glossary
By AnyCap Team
Shared language
for agent systems.
Use these glossary entries to align teams on the terms behind capability runtimes, harnesses, and context design.
Why this glossary exists
Agent systems often fail at the language layer before they fail at the tooling layer. Teams use the same words to mean different things, which turns design discussions into confusion around the model, harness, runtime, or protocol layer.
The glossary gives AnyCap pages a shared reference frame. It helps readers connect conceptual terms such as context engineering, agent harnesses, and capability runtimes to the more practical pages elsewhere in the site.
Context Engineering
The practice of shaping inputs, files, and memory so an agent can work reliably.
Agent Harness
The execution environment that manages tools, files, prompts, and agent safety rails.
Agent Capability Runtime
A runtime layer that gives agents a stable way to call capabilities across providers.
Model Context Protocol
The open communication standard that defines how AI agents discover and invoke tools.
Agentic Workflow
A goal-directed sequence of AI decisions and tool actions that runs without per-step human approval.
What you will find here
- Concise definitions for recurring agent-system concepts.
- Terminology that links conceptual pages to setup guides and comparison pages.
- A cleaner vocabulary for deciding where AnyCap fits in the broader agent stack.