Glossary

An agent control plane is the infrastructure layer that inventories, governs, orchestrates, and provides observability across an organization's fleet of AI agents — regardless of which framework or vendor built them.

The marketplace and catalog for discovering and installing open-source, community-built agents to run on Guild.

Agent sprawl is the uncontrolled proliferation of AI agents across an organization without centralized visibility, governance, or ownership — the AI equivalent of shadow IT, but faster and with deeper system access.

The three lifecycle stages of an agent version: Saved (code uploaded), Validated (build passed), Published (available to the org). Agents must be published to be visible to teammates.

The Agent-to-Agent Protocol (A2A) is an open standard developed by Google that enables AI agents to discover, communicate with, and delegate tasks to other agents—forming the networking layer for multi-agent systems.

Agentic AI describes AI systems designed to operate with autonomy—perceiving their environment, making decisions, and taking multi-step actions to achieve goals without constant human intervention.

An agentic workflow is a goal-driven process where AI agents autonomously plan, execute, and adapt multi-step tasks with minimal human intervention — unlike traditional automation that follows rigid, predefined rules.

AI agent governance is the runtime enforcement of policy, identity, and audit on AI agents in production — distinct from agent monitoring and from human access controls.

AI agent observability is the practice of instrumenting, tracing, and analyzing the end-to-end behavior of autonomous AI agents — including their reasoning steps, tool invocations, inter-agent communication, and decision outcomes — so engineering teams can debug, govern, and improve agents in production.

AI Agent Orchestration is the practice of coordinating multiple AI agents to work together on complex tasks—managing task decomposition, agent selection, execution sequencing, and result aggregation across multi-agent systems.

AI agent portability is the ability to move an AI agent — its logic, configuration, memory, tool integrations, and behavioral context — across different models, frameworks, platforms, and infrastructure environments without rebuilding from scratch.

An AI agent runtime is the execution environment that manages how AI agents run, persist state, recover from failures, and interact with external systems in production.

AI agent testing is the process of evaluating autonomous or semi-autonomous AI systems to ensure they perform tasks correctly, safely, and reliably across real-world conditions.

AI coding assistants are intelligent software tools that use large language models to help developers write, refactor, test, and understand code faster through context-aware support directly inside their IDE.

AI developer productivity measures how much real value engineers deliver using AI tools across coding, testing, documentation, and operations—capturing speed, quality, collaboration, and business impact beyond raw output.

AI developer tools use machine intelligence to accelerate software creation by automating repetitive tasks, generating context-aware code, improving debugging, and enhancing collaboration. Developers using these tools complete 26% more tasks weekly, document code in half the time, and report higher satisfaction — making AI tools essential to modern, efficient engineering workflows.

AI evaluation (evals) is the structured process of testing and measuring the performance, accuracy, and reliability of AI systems — including LLMs, RAG pipelines, and autonomous agents — against predefined metrics and business objectives.

AI guardrails are the technical controls, validation layers, and policy enforcement mechanisms that constrain AI system behavior within defined safety, compliance, and operational boundaries.

AI hallucinations occur when generative models produce false or misleading outputs with high confidence. They arise from statistical prediction errors rather than intentional deception.

An AI IDE (Artificial Intelligence Integrated Development Environment) is an intelligent coding platform that uses machine learning to automate development tasks, generate code from natural language, and improve productivity and code quality.

AI pair programming is a software development practice in which a human developer collaborates in real time with an AI assistant that suggests, generates, reviews, and explains code within the developer's workflow.

API throttling controls how many requests a client can send to an API within a set time window, protecting backend systems from overload and ensuring fair, stable access for all users.

Brownfield AI adoption is the practice of deploying artificial intelligence systems — including AI agents, ML models, and intelligent automation — into enterprise environments where legacy infrastructure, existing data pipelines, and established business processes are already in production.

The ability to bring an agent created elsewhere on a different platform to Guild.

Code review automation uses software tools and AI systems to analyze source code for bugs, vulnerabilities, and coding standard violations, providing faster, consistent, and scalable feedback before human review.

Coding efficiency is the practice of writing software that accomplishes its goals using minimal time, memory, and compute resources — improving performance, scalability, cost, and maintainability without sacrificing functionality.

Data residency is the geographic location where an organization’s data is stored and processed, shaping how laws apply to that data, as well as compliance, security, performance, and market access.

Developer Experience (DX) is the overall quality of interactions, perceptions, and friction points developers encounter while building, testing, deploying, and maintaining software.

Developer friction is the cumulative tax of everything that prevents engineers from writing code — context switching, brittle tooling, unclear documentation, slow CI/CD pipelines, and organizational inefficiencies that compound into lost days every week.

An embedding model converts unstructured data—such as text, images, or audio—into dense numerical vectors that preserve semantic meaning, enabling AI systems to understand relationships and similarity between items.

A temporary agent version created during guild agent test or guild agent chat that does not appear in version history. Unchanged files reuse a cached ephemeral build.

Fine-tuning rewrites a model’s internal knowledge, while prompt engineering shapes its behavior through structured inputs. Together, they give developers two complementary levers for customizing LLMs.

Function calling is the ability of a large language model to analyze a user's natural language input, determine that an external function or API should be invoked, and produce a structured output specifying the function name and arguments — without executing the function itself.

A General AI Agent is an autonomous software system that perceives its environment, reasons about goals, and takes actions across multiple domains without continuous human oversight.

Block's open-source AI agent framework. Guild supports agents defined as Goose recipe.yaml files.

Guild Insights provides developers and organizations with detailed visibility into how their AI agents consume resources and incur costs. Insights helps teams identify where to focus on making agents more efficient. Knowing what an agent is spending is the first step to knowing what to improve.

An agent type defined using only Markdown files (PROMPT.md, DESCRIPTION.md) with no TypeScript. Skips build validation entirely.

Human-in-the-loop (HITL) is a system design pattern in which a human actively participates in the operation, supervision, or decision-making of an automated or AI-driven process, typically by approving, rejecting, or modifying outputs at critical checkpoints before they take effect.

Interoperability is the ability of different software systems, tools, and services to exchange and use data seamlessly, enabling connected workflows, reducing silos, and allowing developers to integrate diverse technologies without manual effort or custom workarounds.

A Python-originated text templating format used in Goose recipe files for instructions and prompt fields.

JSON Schema in AI is a declarative specification that constrains large language model outputs and tool interactions to predictable, machine-readable structures — ensuring that every response, function call, or agent action conforms to a developer-defined contract.

A knowledge graph is a structured data model that represents real-world entities — people, places, events, concepts — and the relationships between them as a network of nodes and edges, enabling machines to query, reason over, and infer new facts from connected information.

LLM observability is the practice of collecting, tracing, and analyzing telemetry data from large language model applications in production to understand not just if something is wrong, but why, where, and how to fix it.

Agents can express ordered preferences for LLM providers and models, but org-level Model Policies have final authority. Preferences are hints; policies are enforced rules.

Low-Rank Adaptation (LoRA) is a parameter-efficient fine-tuning method that freezes a model’s original weights and trains only small low-rank matrices inserted into each layer—cutting trainable parameters by up to 10,000× while matching full fine-tuning performance.

The Model Context Protocol (MCP) is an open standard that defines how AI agents connect to external tools, APIs, and data sources—acting as a universal interface that lets any AI system plug into any compatible resource without custom integration work.

A monorepo stores multiple independent projects in a single repository, enabling unified dependencies, streamlined CI/CD, shared tooling, and easier collaboration across engineering teams.

Multi-agent systems (MAS) are computational architectures where multiple autonomous agents collaborate, communicate, and coordinate to solve complex problems that single agents or monolithic systems cannot handle on their own.

A non-deterministic system is a system that can produce different outputs from the same input under identical conditions, meaning its behavior is not fully prescribed by its specification and cannot be exactly repeated across runs.

On-premise AI refers to artificial intelligence models and infrastructure deployed entirely within an organization’s own environment—offering maximum control, security, and performance without relying on external cloud providers.

A standard machine-readable format for describing REST APIs. Referenced in Guild as a way to bulk-import integration operations via guild integration operation create --openapi ./openapi.yaml.

Prompt engineering is the process of designing, testing, and optimizing natural language inputs — called prompts — to reliably elicit desired outputs from large language models (LLMs) and other generative AI systems.

Guild's global naming convention uses a tilde (~) as a separator between owner and resource name (for example, guildai~the-smith, myorg~my-agent).

Query routing is the process of analyzing an incoming request and directing it to the most appropriate model, data source, index, or processing pipeline based on the query's intent, complexity, and operational constraints.

Retrieval-Augmented Generation (RAG) is an architecture pattern that enhances AI models by retrieving relevant information from external sources before generating responses—grounding outputs in current, specific, and verifiable data.

Semantic search retrieves information by understanding meaning and intent — not just matching keywords — using NLP, vector embeddings, and machine learning to deliver far more accurate, intuitive, and context-aware results.

Semantic understanding enables AI systems to interpret meaning, intent, and context — not just literal keywords. It powers semantic search, conversational AI, developer tools, and any system that needs to understand natural language the way humans do.

A version-constraint syntax (for example, ^1.4.0) used when Guild artifacts declare dependencies on other versioned artifacts. The caret means "compatible minor versions," following the npm convention.

A full log of an agent's execution run, including events and state. Sessions can be resumed with --resume <session-id>. Distinct from a login session: a poll timeout leaves a session intact rather than discarding it.

Reusable content packages that inject Markdown instructions into an agent's context at runtime based on description relevance. The agent sees only a skill's title and description and decides when to pull in the full body content. A skill is a way to share knowledge across multiple agents: your organization's writing style, a policy for managing labels on your GitHub issues, and so on.

Guild's built-in support and assistant agent, opened via guild chat.

Specialized AI agents are autonomous systems designed to perform specific tasks or operate within defined domains, delivering higher accuracy and efficiency than general-purpose AI through targeted training and focused capabilities.

The practice of aggressively maximizing the consumption of AI tokens, the basic units of text processed by language models. It is used as a vanity metric for productivity and an optimization strategy for workflow automation.

Tool use in AI agents is the capability that allows a language model to invoke external functions, APIs, databases, or services to complete tasks it cannot perform with its training data alone.

Guild's mechanism for automating agent execution. Supports webhook triggers (fired by external service events, e.g., a Slack app_mention) and time-based triggers (cron-like scheduling).

Unit testing for AI agents is the practice of validating the individual components of an AI agent system — including prompt templates, tool-calling logic, memory operations, and reasoning chains — in isolation to verify they behave correctly before the agent runs as a complete system.

A vector database stores and searches high-dimensional embeddings to find similar items based on meaning rather than exact matches, powering semantic search, recommendations, and retrieval-augmented AI systems.

Webhook automation is the practice of using event-driven HTTP callbacks to trigger automated workflows between systems the moment a relevant event occurs.

Guild's primary organizational container, holding agents, triggers, context, and credential policies. Not a folder or a team, closer to a deployment environment.

Versioned Markdown content attached to a workspace and prepended to every agent session. Has its own draft/publish lifecycle. Distinct from an LLM's context window.

XGBoost (eXtreme Gradient Boosting) is a popular and powerful machine learning algorithm used for supervised learning tasks, particularly in regression, classification, and ranking problems.

YAML configuration in AI systems is the practice of using YAML (YAML Ain't Markup Language) — a human-readable data serialization format — to declaratively define the parameters, behaviors, and infrastructure of machine learning pipelines, AI agents, and intelligent workflows.

Zero-shot learning is a machine learning paradigm in which a model classifies or performs tasks on categories and inputs it was never explicitly trained on, using transferred semantic knowledge instead of labeled examples.

A TypeScript runtime schema validation library used throughout the Guild SDK to define inputSchema and outputSchema for agents and tools.