Glossary

Every term this guide uses, in one place.

Word	Meaning here
Instance	One running copy serving one organisation: a config file, a database, a set of users.
Agent	A named AI persona with its own instructions, model, safety mode and granted tools.
Engine	The thing that actually runs the conversation with a model. Three are interchangeable.
Engine seam	The adapter boundary that keeps provider-specific types out of the rest of the codebase.
Connector	A broad ability switched on or off for a whole instance.
Capability	A specific external tool an operator added and granted to one agent.
Resource	A repository plus a written guide an agent is allowed to see.
Workspace	The cached repository copies, with one isolated checkout per agent.
Seat	A licence for one actor — human or agent — to act on an instance.
Gate	The pause before a tool runs, where the rules are checked.
Safety mode	An agent’s coarse setting: `readonly`, `approval-gated`, or `readwrite`.
Tool policy	A per-tool rule that can tighten a single ability without demoting the whole agent.
Separation of duties	The requester cannot be the approver — enforced by the permission engine.
Coding controller	The pipeline that turns an issue into a reviewable merge request and stops there.
Idempotent	Running it twice produces the same result — an updated merge request, not a second one.
Constellation	One instance driving other instances and desktop nodes.
Interrupt governor	Decides whether an agent’s unprompted output interrupts you or waits in an inbox.
Audit log	The permanent record of who asked, who decided, what ran, and what it cost.
MCP	The standard interface other programs use to talk to neutron, and neutron uses to talk to tools.
Persona plugin	Optional per-instance branding and behaviour, shipped separately from the core image.
Schedule	A cron rule that starts an agent turn with no human present.
assay-engine	The sidecar binary providing the encrypted vault, cron, and durable workflows.