Glossary

Workspace

A local project container in Sentinel. Usually backed by a folder on disk. Threads belong to a workspace.

Thread

The main unit of work in the app. A thread stores messages, mode, engine, model state, queued follow-ups, and repo-related state.

Chat mode

The default conversational work mode for a thread.

Plan mode

A structured thread mode with a stored plan document, tracked tasks, and question sets.

Engine

The runtime layer used for a thread. Sentinel supports sentinel, codex, claude, and copilot.

Provider

The credential and connection layer for AI model access, such as OpenAI or Anthropic.

Model

A concrete model ID used through an engine and provider setup.

Worktree

A repo checkout path associated with a thread so work can stay isolated from the main local checkout.

Checkpoint

A saved repo state tied to thread history so earlier work can be restored or branched from.

Approval

A confirmation step required before a tool or group of tools can run.

Approval policy

The rule set attached to a thread runtime that decides when the runtime must stop and ask before acting.

Sandbox mode

The access level a runtime gets when it is acting on the local machine.

Skill

An installable capability pack that extends Sentinel, Claude, or Codex behavior.

MCP server

A Model Context Protocol server connected to Sentinel over STDIO or HTTP.

Integration

A built-in external system connection such as GitHub, Gmail, or PostgreSQL.

Automation

A recurring prompt tied to a workspace and runtime configuration that runs on a schedule.

Sub-agent

A delegated run linked back to a parent thread and stored with its own child state.

Project mode

The repo mode a thread is using, usually the main project checkout or a thread worktree.

Generated media

Image or video output produced through the app's media generation paths and served back to the UI as temporary artifacts.