| Activity | A unit of work in a program phase, carried out by a workflow, agent, team or person. See Program activities. |
| Agent | A model plus instructions, functions (tools), knowledge and memory, built in Build → Agents. See Agents. |
| AI system | A registered grouping of agents and teams that serve one business purpose, with an owner, lifecycle and impact assessment. See AI systems. |
| API key | A credential for calling the FYAI API, created in Settings → API Keys and bound to one project with Run, Read, Write or Admin access. See Authentication. |
| Approval | A pause before a tool call or workflow step until someone approves or rejects it. See Approvals and risk appetite. |
| Approval Override | A per-team, per-agent or per-function setting that pins approval to always or never required, overriding the risk appetite. See Using approval gates. |
| Assurance pack | An exportable bundle of governance evidence for a period. See Assurance pack. |
| Attribute | An input defined on a workflow and filled in when a mission is launched. See Workflow attributes. |
| Blueprint | The platform-level original of a template. Your project receives copies of it. See Templates. |
| Built-in tool | A tool the platform gives agents automatically, such as remember, recall or hand_off_to_agent. See Built-in tools. |
| Chat drawer | The chat panel you can open from any page, docked or floating. See Chat drawer. |
| Checkpoint | A sign-off that holds a program phase until it's approved. See Programs. |
| Concierge | The platform's guide agent, which opens in chat for new users and helps you find and build things. |
| Connected account | Your own sign-in to an external service (for example Atlassian or Microsoft 365) that agents use when acting for you. See Connected accounts. |
| Containment | An MCP Gateway control that quarantines or blocks an identity, server or tenant. See MCP Gateway. |
| Control library | Catalogues of governance controls with the evidence that shows whether each one holds. See Control library. |
| Dashboard | A grid of tiles fed by live data sources. See Dashboards. |
| Data source | A refreshable feed of data from an API, database, internal query or event stream. See Data sources. |
| Delegation | Handing a task to the platform from an external tool such as Claude Code, and tracking it to completion. See Delegating from Claude Code. |
| Element | The definition a dashboard tile is rendered from, such as a chart, table or custom HTML. See Elements. |
| Event source | A WebSocket or server-sent-events stream the platform listens to. See Event sources. |
| Execution canvas | The live view of a mission, showing each step's status. See Missions. |
| Execution environment | A machine where agents' tools run, usually your own computer or server connected with the runner. See Execution environments. |
| Function | A tool an agent can call: your own code, an MCP server, an integration operation, a runner tool and more. See Functions. |
| Hand-off | Passing a conversation from one agent to another. See Chat approvals and hand-offs. |
| Impact assessment | A structured assessment of an AI system's risks, required before it goes live. See AI systems. |
| Integration | Stored connection details and secrets for an external service, used by functions, models and data sources. See Integrations. |
| Killswitch | A control that immediately stops an agent from running. See Observability. |
| Knowledge base | A searchable collection of documents that agents query. See Knowledge bases. |
| Manifest | A structured document, validated against a schema, that agents and features read as configuration or reference data. See Manifests. |
| MCP | Model Context Protocol, the standard for connecting agents to tool servers. FYAI can use MCP servers as functions, expose tools through the MCP Gateway, and be driven from the platform MCP server. |
| MCP Gateway | The governed endpoint through which external agents and clients use MCP tools. See MCP Gateway. |
| Memory | What agents remember across conversations: facts extracted from sessions and notes agents save themselves. See Memory. |
| Mission | One run of a workflow. See Missions. |
| Model | An LLM configuration agents use, either platform-provided or your own key. See Models. |
| Model group | Several models behind one name, with routing between them. See Model groups. |
| Organisation | The top-level account that owns projects, members, billing and governance settings. See Organisations and projects. |
| Phase | A stage of a program, made up of activities. See Programs. |
| Platform-managed | A read-only copy of a platform template in your project. You can use it but not edit it. See Templates. |
| Program | Long-running, multi-phase work with activities and checkpoints. See Programs. |
| Project | A workspace inside an organisation. Most things you build belong to one project. See Organisations and projects. |
| Project role | Your level of access in a project: Admin, Editor, Viewer or Runner. See Members and roles. |
| Promotion | Copying a team or workflow, with what it depends on, from one project to another. See Promoting between projects. |
| Reason code | The recorded reason a policy decision allowed, required approval for, or denied an action. See Reason codes. |
| Review queue | Where risk ratings the platform proposed wait for a person to confirm them. See Review queue. |
| Risk appetite | How much risk your organisation, project or you accept before an approval is required. See Approvals and risk appetite. |
| Risk level | The classification of an agent, team or function from informational to critical. See Risk classification. |
| Run grant | Permission for a project member or key to launch a specific workflow or agent. See Members and roles. |
| Runner | The FYAI runner program you install on your own machine to connect it as an execution environment. See Runner. |
| Schedule | A timetable that launches a workflow automatically. See Scheduling workflows. |
| Service account | A non-human identity for programmatic access, using a client ID and secret. See Authentication. |
| Stack | An installable bundle of agents, teams, workflows, dashboards and other components from the Marketplace. See Marketplace. |
| System message | The instructions that define an agent's role and behaviour. See Agents. |
| Team | Several agents working together under one orchestration pattern: Sequential, Concurrent, Group Chat, Group Chat Selector, Handoff or Magentic. See Teams. |
| Template | A platform-provided agent, team, function, model, workflow or other item that is copied into projects. See Templates. |
| Test case | An input and expected outcome used to evaluate an agent or team. See Testing agents and teams. |
| Token | The unit LLMs process text in. Usage and cost are measured in tokens. See Usage and billing. |
| Webhook | An HTTP endpoint that starts a workflow when an external service sends an event. See Webhooks. |
| Workflow | A graph of steps — agents, teams, functions, conditions, loops and approvals — that runs the same way every time. See Workflows. |