The three names
AMAS
Artaios Multi Agent System. The declarative agent system: the YAML
configuration language, the runtime that executes it, and the REST and
WebSocket APIs.This is what these docs are about.
Agent Studio
AMAS’s web interface at amas.artaios.com. Where
you write YAML, test in the Playground, build knowledge bases and manage
keys.
Artaios
The end-user chat product. Not a build surface — a place your finished
agents can be published to so non-technical colleagues can use them.
How they relate
Which word to expect where
Agent Studio is not a separate product from AMAS. It is AMAS’s UI. Anything
you can do in Agent Studio you can also do over the API — the YAML is the same
YAML either way.
Glossary
Agent
Agent
A single configured unit with a model, a prompt, and optionally tools,
knowledge and memory. AMAS has two types: LLM agents (conversational)
and ReAct agents (tool-using, reasoning). See
Agent Types.
Agentic system
Agentic system
One or more agents plus the orchestration between them, defined in a single
YAML config. A single-agent config is still an agentic system. See
Multi-Agent Introduction.
YAML config
YAML config
The declarative source of truth for an agentic system. Hand-written,
reviewable, version-controllable. This is the primary interface to AMAS —
see YAML Configuration.
Playground
Playground
The Agent Studio page where you edit YAML and run the resulting system
interactively. See Playground.
Agent key
Agent key
The identifier for a saved, deployed agent. Used to open a session over the
REST or WebSocket API, and to point the embedded chatbot at an agent.
Session
Session
One conversation with an agentic system, holding its message history and
state. Created from a YAML config or from a saved agent key.
Knowledge base
Knowledge base
A set of file collections, indexed for semantic search, that agents retrieve
from at runtime. See
Knowledge Base Setup.
Toolkit
Toolkit
A grouped set of tools for one external service (Outlook, Teams, Jira, …),
sharing one authorization. See Tools.
HITL (Human-in-the-Loop)
HITL (Human-in-the-Loop)
Approval gating: a consequential tool call pauses and waits for a human
before it runs. See
Human-in-the-Loop.
BYOK (Bring Your Own Key)
BYOK (Bring Your Own Key)
Using your own LLM provider API keys instead of the pooled ones, for direct
control over usage, billing and rate limits. See
Model Selection.
Next
Quickstart
Build and run your first agentic system
YAML Configuration
The full config reference