Glossary
A comprehensive list of terms and definitions used throughout AgentGraph.
A
Agent: An autonomous software program designed to perform tasks, make decisions, or interact with users or other systems, often powered by AI models.
Agent Chat: The in-app AI assistant available to Buyers and Builders. It can answer platform questions, search these docs, help Buyers design Agent Requests and find engineers, and help Builders manage their profile and availability.
Agent Request: A request by a Buyer to have an Agent built for them. Agent Requests include the agent's requirements, goals, integrations, data sources, budget, and success criteria to provide Builders with adequate information to submit a proposal.
AI Engineer: The Buyer-facing term for a Builder — an elite AI-native engineer who builds agents on AgentGraph. See AI Engineer Types.
B
Bid (Proposal): A proposal submitted by a Builder in response to an Agent Request, covering their hourly rate, expected weekly hours, approach, and timeline. "Bid" and "proposal" are used interchangeably.
Builder: A developer who creates AI agents on the AgentGraph platform, typically an elite AI engineer fluent in the latest agent frameworks and code-generation tooling.
Build Room: The shared workspace created for an Agent Request once a Builder is selected. It covers the whole engagement: terms, proposals, payments and hours tracking, delivery review, and reviews.
Buyer: An individual or organization that commissions the development of custom AI agents through AgentGraph.
C
Certification Level: A Builder's verification status on AgentGraph: Not Certified, GitHub Verified (earned by connecting GitHub), or AgentGraph Elite (awarded by AgentGraph).
D
Data Room: A Buyer organization's file repository on AgentGraph for sharing documents and context with Builders.
E
Engagement: An active working relationship between a Buyer and a Builder — either a scoped project or an ongoing hourly retainer, governed by the agreed engagement terms (rate, weekly hours, scope).
G
GitHub: A web-based platform for version control and collaborative software development using Git. Used to host, review, and manage source code for agent projects, and to verify Builder history on AgentGraph.
H
Hosting: The process of deploying and running an agent or application on a server or cloud infrastructure, making it accessible to users or other systems.
Human in the Loop: An approach in AI and automation where human feedback, oversight, or intervention is integrated into the workflow to guide, validate, or improve the system's outputs and ensure alignment with user goals.
I
Inference: The process of using a trained AI model to make predictions, generate outputs, or perform tasks based on new input data.
Integration: The process of connecting an agent or application to external systems, APIs, or data sources to extend its functionality.
M
Marketplace: Where supply meets demand on AgentGraph. Buyers browse and invite AI Engineers or book Services; Builders browse open Agent Requests to propose on.
Model: An AI system trained on data to perform specific tasks such as classification, generation, or decision-making. Models are the core components powering agents' intelligence.
Modality: The type of data or input that an AI model is designed to process. Common model modalities include text, image, audio, video, and multimodal (supporting multiple types). The modality of a model impacts the scope and capabilities of an agent—text models handle language, image models process visual data, audio models interpret sound, video models analyze moving images, and multimodal models can combine several types for richer, more complex tasks.
MCP (Model Context Protocol): An open protocol that lets AI assistants securely connect to external tools and data sources through a consistent interface. AgentGraph exposes its own MCP connector so you can use the platform from Claude and other MCP clients.
P
Platform User Agreement: The agreement governing use of AgentGraph, accepted in-app before transacting. See Legal.
R
RAG (Retrieval-Augmented Generation): An AI technique that combines information retrieval from external sources with generative models to produce more accurate and contextually relevant outputs.
Retainer: An ongoing hourly engagement with a Builder, billed weekly from logged and approved hours — the standard way to continue work after delivery or run open-ended AI engineering capacity.
S
Scoped Project: An engagement with a defined agent build and agreed success criteria, delivered and accepted in the Build Room.
Service: A pre-scoped, fixed-price offering listed in the Marketplace that Buyers can book directly, delivered by AgentGraph Elite Builders.
Source Code: The human-readable instructions and files that make up a software program or agent, typically written in programming languages and managed in version control systems like Git.
Statement of Work (SOW): The document that captures a specific engagement's scope, rates, and terms between a Buyer and Builder(s), operating under the Platform User Agreement.
Success Criteria: The specific, agreed-upon requirements and measurable outcomes that an agent must achieve to be considered successfully delivered. Success criteria are defined during the Agent Request and Build Room processes and are used to guide development, testing, and final approval.
T
Test Case: A specific scenario or set of conditions used to verify that an agent or software system behaves as expected and meets its requirements.
Testing: The process of verifying that an agent or software system functions correctly and meets its specified requirements through systematic evaluation and validation.
W
Weekly Hours: The billing unit for hourly engagements and retainers: the Builder's logged daily hours and notes for a week, submitted for Buyer approval and then invoiced. See Payments.