Glossary
Agentic AI, defined plainly.
The terms you'll meet when evaluating AI agents, MCP and natural-language analytics, each in two sentences or fewer.
Agentic AI
- AI systems that plan and take actions toward a goal, such as calling tools, updating records or running code, rather than only generating text in reply to a prompt. Learn more
Agent harness
- The runtime around a language model that supplies tools, memory and routing, and enforces permissions, budgets, verification and audit so the model can work as a dependable agent. Learn more
Model Context Protocol (MCP)
- An open protocol that lets AI applications discover and call tools and data sources exposed by MCP servers through a standard interface. Learn more
MCP gateway
- A single control point between agents and MCP servers that checks, brokers, meters and records every tool call. Insiro's gateway runs 13 checks per call. Learn more
Capability token
- A short-lived credential that lets one agent task call specific tools with specific arguments. In Insiro it is bound to tenant, user, mission and attempt, and revocable. Learn more
Risk tier
- A classification of how dangerous an action is. Insiro uses tiers 0 to 3; tier 2 requires user confirmation and tier 3 requires owner approval. Learn more
NL to SQL
- Natural-language-to-SQL: translating a question in plain language into a database query. Datasight does this through a 10-step validated pipeline. Learn more
Semantic layer (ontology)
- A business description of data: tables, metrics, dimensions, relationships and sensitive fields. Datasight discovers it automatically into editable YAML. Learn more
Federated query
- A single question answered using data from more than one database, each queried in its own SQL dialect. Learn more
AI teammate
- A persistent AI agent with a role, persona, model, budget and app ceiling that can lead a goal and build a team of sub-agents. Learn more
Sub-agent
- A specialised agent created by a lead agent for part of a goal. In Insiro a sub-agent can never use apps beyond its lead teammate's ceiling. Learn more
Acceptance criteria
- Conditions that define when a goal is done. Insiro checks a goal's result against each one and marks it met or not met. Learn more
Routine
- A standing goal that watches for a condition or schedule and starts the right pipeline or goal when it occurs. Learn more
Skill
- A reusable procedure taught by demonstration in Insiro's cloud browser and shared like a document. Learn more
DLP
- Data loss prevention: scanning content for sensitive data and blocking or redacting it. Insiro applies DLP to tool inputs and outputs. Learn more
BYOK / BYOC
- Bring your own key (use your own model-provider API keys) and bring your own cloud (deploy Insiro in your own Azure subscription). Learn more
See Insiro on your own stack.
A 30-minute working session: we connect one data source and one app, then run a governed agent on a task you choose.
- 13governance checks on every agent tool call
- 1,700+SaaS apps agents can act in
- 30+data dialects for plain-English questions