Skip to content

Glossary

Platform Terms in Alphabetical Order

  • Access level: How much a role lets the holder do with a resource, from Read-only to Full access. See Roles and teams.
  • Agent: A worker that you chat with and that runs workflows. Each agent uses exactly one sandbox. See Agents and sessions.
  • Agent capability: A workspace permission that controls who creates, shares and uses agents, such as Author or Use Shared. See Roles and teams.
  • Allowed host: A domain or address range that a sandbox lets its agents reach. See Allowed hosts.
  • API key: A key you create to call agents or workflow webhooks. See API keys.
  • Coding workspace: A workspace type, marked Preview. It adds GitHub projects and Git worktrees and has no workflows or dashboards. See Organizations and workspaces.
  • Consent required: A setting on an MCP tool in a sandbox. Chat asks you to Deny, Always allow or Allow once. See MCP connections.
  • Custom tool: A Bash, Python or Node.js script that an agent can call as a tool. See Custom tools.
  • Dashboard: A page of charts and tables that an agent builds and keeps fed with data. See Dashboards.
  • Edge: An arrow from one node of a workflow to the next. See Workflows.
  • Evaluation: A test that runs one workflow on several models and scores each run with a judge model. See Evaluations.
  • Event Trail: A 30-day log of admin changes. See Audit and observability.
  • General purpose: The workspace type with agents, workflows, triggers and dashboards. See Organizations and workspaces.
  • Immutable skill: A skill that you upload. It has versions. See Skills.
  • Inference gateway: The AgentZ service that carries model calls from agents to your provider. See Security model.
  • Inference pool: A workspace list of models in order. One model name routes through it. See Inference providers.
  • Inference provider: A connection to a model source, such as OpenAI or your own endpoint. See Inference providers.
  • Lens: The workspace pages for traces, runtime telemetry and MCP activity. See Audit and observability.
  • MCP connection: A link to an outside tool server. See MCP connections.
  • Mutable skill: A skill that lives in one agent's home directory. The agent can edit it. See Skills.
  • Node: One step of a workflow. See Workflows.
  • Organization: The top-level container for members, roles, teams and workspaces. AgentZ creates one at sign-up. See Organizations and workspaces.
  • Package: A Nix package installed for the agents of a sandbox. See Sandbox packages.
  • Permission scope: The place where a role's permissions apply, either the organization or one named workspace. See Roles and teams.
  • Persistent memory: A switch on an agent that saves facts and journal entries across sessions. See Agents and sessions.
  • Role: A named set of permissions that you give to a person or a team. See Roles and teams.
  • Sandbox: A reusable configuration of models, packages, MCP tools, skills and allowed hosts. It is not a separate machine. See Sandboxes.
  • Schedule: A cron timetable that starts a workflow run. See Schedules.
  • Secret: A write-only value that belongs to one agent and works only for the hosts you list. See Secrets.
  • Session: One chat with one agent. A workflow run also starts one. See Agents and sessions.
  • Skill: A folder with a SKILL.md file that teaches an agent a repeatable job. See Skills.
  • Social admission: An organization page that lets people join with a matching Google or GitHub account. See Users and invitations.
  • Span: One step inside a trace, such as a model call or a tool call. See Audit and observability.
  • Superadmin: The built-in role with every capability in the organization. See Roles and teams.
  • Team: A group of members that share roles. See Roles and teams.
  • Trace: The record that Lens shows for agent activity. It is made of spans. See Audit and observability.
  • Trigger: A schedule or a webhook that starts a workflow run. See Schedules and Webhooks.
  • Webhook: A URL call that starts a workflow run, authorized by a Webhook API key. See Webhooks.
  • Widget: One chart or table on a dashboard. See Dashboards.
  • Workflow: A saved graph of nodes that an agent creates in chat. See Workflows.
  • Workflow run: One execution of a workflow, with a status such as Running or Succeeded. See Workflows.
  • Workspace: A place where a team builds and runs agents. See Organizations and workspaces.
  • Workspace Admin: The built-in role with every capability in one workspace. See Roles and teams.
  • Zero secret exposure: The agent holds a placeholder in place of the secret value. AgentZ adds the real value to requests for the hosts you list. See Security model.

Next Step

Continue with Limits.