Glossary
Definitions of the terms used throughout the AlphaAgent documentation. Where a concept has its own chapter, the definition links to it.
Active configuration
The setting that determines which version of an agent or swarm is currently live, along with the knowledge graph, connectors, and execution environment that the live agent uses. Editing an agent creates a new version but does not change what is running until you activate it. See Agents.
Agent
A versioned worker you define — a specialist focused on a domain or a supervisor that orchestrates a swarm — with its own instructions, model settings, and connected data. Every edit creates a new version. See Agents.
AMPG (Augmented Multi-Resolution Property Graph)
The product name for an AlphaAgent Knowledge Graph — a graph built from your documents that agents can retrieve from. It is multi-resolution (your document is structured at several levels of detail at once) and retrieves by meaning rather than keywords, re-ranking results by the relationship a question is about. See Knowledge Graphs (AMPG) and Knowledge graphs in depth.
Coder
The spec-driven code component, built on the Claude Agent SDK, that agents delegate analysis to. It runs in Studio's code-interpreter service, executes code in your sandbox (the execution environment), reads attached documents, fetches connector data, and writes durable artifacts (datasets, charts, docs, reports) into the conversation's S3 store. See Code interpreter and datasets.
Console
The web portal, hosted by Prometheus Research Labs, where you manage your account, organization, billing, license keys, and Studio downloads. The Console never connects to your AWS account. See Console overview.
Data connector
A versioned link from an agent to an external data source — a SQL database, REST API, external tool (MCP server), or AWS data source. See Data connectors.
Dataset
A result table that an agent registers during a conversation. Instead of dumping every row into the chat, the dataset is kept in your workspace and the chat shows a compact preview, chart, or grid from it. See Datasets and data flow.
Deep Traversal
A chat mode that plans before acting, draws on knowledge-graph retrieval, and can present a plan for your approval before carrying out deeper, multi-step work. Every Deep Traversal flow starts from AMPG grounding, so it requires a Knowledge Graph: an agent without one can't run Deep Traversal and should use Express instead. Contrast with Express. See Chat.
Execution environment
The sandboxed compute where an agent runs code, with configurable memory, storage, and time limits. See Execution environments.
Express
The fast, direct chat mode for straightforward questions and quick tasks. Contrast with Deep Traversal. See Chat.
Handoff
A workflow node type that pauses a run to wait for human input, then resumes once a person responds. See Workflows.
Heartbeat
A periodic signed message Studio sends to the Console to confirm the license is healthy. See Licensing and telemetry.
Knowledge Graph
See AMPG.
License key
The entitlement, created in the Console, that authorizes one Studio deployment. Its public identifier starts with aalk_. See License keys.
License token
The one-time secret value shown once when you create a license key, in the form aalk_<id>.<secret>. You provide it during deployment so Studio can activate the license. It is shown only once — copy it immediately. See License keys.
Multi-resolution graph
A property of an AMPG: your document is represented at several levels of detail at once — whole document, sections, passages, and individual phrases — so retrieval can move between broad context and a single decisive line. See Knowledge graphs in depth.
Neo4j (self-hosted)
How your knowledge-graph database is provided. Studio runs single-node Neo4j Community as a Fargate service in your VPC, backed by encrypted multi-AZ EFS (zero-touch, no Marketplace subscription). The graph stays in your AWS account. See Prerequisites.
Organization
Your tenant in the Console: the account that owns your billing, licenses, and team members. See Organization and members.
Relation-aware retrieval
How an AMPG finds answers: a broad, meaning-based pass gathers candidate passages (matching paraphrases, synonyms, and terms not seen at ingestion time), then a precision pass re-ranks them by the specific relationship the question is about. See Knowledge graphs in depth.
Specialist
An agent that acts as a domain worker inside a swarm; the supervisor routes sub-tasks to it. In a swarm chat you can also direct a question to a specialist directly. See Swarms.
Studio
The AlphaAgent application your team deploys into your own AWS account and works in day to day. See What runs in your account.
studioctl
The guided command-line tool, shipped in the Studio bundle, that manages the full deployment lifecycle in your AWS account. Its subcommands include install, update, reconfigure, preflight, status, doctor, maintenance, saml, and teardown. It is resumable and idempotent — an interrupted run can be re-run and continues from where it stopped. It supersedes the older deploy.py installer. See Deploy walkthrough.
Supervisor
The coordinating agent of a swarm that receives the request, plans, and delegates to specialists. See Swarms.
Swarm
A team of agents — a supervisor plus specialists — that work together on a task. See Swarms.
Thought Map
A visual map of an agent's reasoning during a run, showing the steps it took as connected nodes. See Chat.
Version (versioning)
AlphaAgent saves a new version of an agent, swarm, connector, or workflow each time you edit it, rather than overwriting the previous one. The active configuration controls which version is live.
Workflow
A multi-step pipeline built as a graph of nodes (task, document-analysis, conditional, and handoff) that can run on demand or on a schedule. See Workflows.
Workspace
The isolated, per-conversation file area (backed by S3) where the coder reads and writes files while running code, organised into datasets, charts, docs, and reports. Uploaded attachments are saved here and the artifacts a run produces are kept here. See Code interpreter and datasets.