---
title: "Glossary"
description: "Carina terms — session, profile, capability, patch, worker, and more."
source: https://carina.nebutra.com/reference/glossary/
---

# Glossary

> Carina terms — session, profile, capability, patch, worker, and more.

Short definitions for terms used across the docs. For product positioning see [Overview](/getting-started/introduction/).

## Session

A durable unit of work bound to a **workspace root** and a **permission profile**. Created with `session.create`. Holds tasks, events, and audit history.

## Workspace

The repository (or directory) tree the session is allowed to touch. Bound at session creation; policy resolves paths against it.

## Profile

A named permission bundle (e.g. `safe-edit`, `read-only`) evaluated by the capability kernel for every side effect.

## Capability

One of the fourteen gated action types (FileRead, CommandExec, PatchApply, MemoryWrite, SubagentSpawn, …). Every meaningful effect maps to a capability check.

## Permission decision

Kernel result: `allow`, `deny`, or `require-approval`. Recorded on the audit trail and referenced by side-effect events.

## Agent loop

ReAct-style cycle: model proposes an action → kernel authorizes → tools execute → observation returns. The model decides; it does not bypass the kernel.

## Sub-agent

An isolated child run with **attenuated** capabilities (never more than the parent). Spawned via SubagentSpawn / `spawn` tooling.

## Worker

A remote or local executor process that polls work under credentials and lease fencing. See [Workers](/deployment/workers/).

## Patch

A transactional file mutation with provenance and a rollback pointer. Preferred write path under `safe-edit` (via PatchApply).

## Audit / event log

Append-only, session-scoped log of decisions and effects. Supports hash-chain verification (`carina audit verify`).

## Item stream

Normalized projection of session activity for UIs and SDKs (distinct from raw audit replay).

## Daemon

Local process (`carina-daemon`) that owns sessions, policy evaluation, and the audit store.

## Gateway

Optional HTTP surface that exposes scoped JSON-RPC / agent APIs when explicitly enabled — not required for local TUI use.

## MCP

Model Context Protocol servers/tools mediated by Carina so external tools still hit policy and audit.

## Nebutra Cloud boundary

Optional identity/sync layer around Carina. **Local machine remains authority** for execution and policy.

## Source of truth

- Product terms: `docs/product.md` · map: `apps/docs/FEATURE_MAP.md`
- Security primitives: `docs/security-model.md` · [Policy](/concepts/policy/) · [Audit](/concepts/audit/)
- Related: [Runtime architecture](/concepts/runtime/) · [CLI reference](/reference/cli/)

## Next

- [Overview](/getting-started/introduction/) — product framing
- [Common workflows](/getting-started/common-workflows/) — recipes using these terms

---
Source: https://carina.nebutra.com/reference/glossary/
Markdown: https://carina.nebutra.com/reference/glossary/index.md