Terms A–Z

A–C

Agent — software that calls a model with instructions; may use tools, memory, and tasks. See Agents track.

AgentFlow — ordered steps: route, parallel, loop. See AgentFlow track.

Approval — human or policy gate before a risky tool runs. See Security — approve a tool run.

D–G

Deploy — serve an agent behind an HTTP endpoint or container. See Deploy track.

Guardrail — code or config that blocks bad output. See Reliability and Security tracks.

Govern — approvals, secrets, and boundaries in production. See lifecycle Govern and Security.

H–M

Job (console) — scheduled or long run in the web dashboard. See Web console track.

MCP — attach an external tool server. See MCP servers track.

Memory — facts saved across turns, not the same as a RAG note. See Memory track.

R–Y

RAG — answer from a section of a note. See RAG track.

Reliability — check and score an answer. See Reliability track.

Skill — instructions loaded when the job matches. See Skills track.

Token — billed unit of model text; see AI basics and Observe tracks.

YAML — agents defined in a file. See YAML track.

Reference: documentation