Glossary
Short definitions of the words used across these docs, with a link to where each one is explained in full.
A
| Term | Meaning |
|---|---|
| Advisor | An opt-in tool that lets the agent ask a second model for advice on a hard decision. Turn it on with /advisor on. See Advisor. |
| Agent | The model working on your task in a loop: it reads, acts with tools, checks the result and decides what to do next. See How it works. |
| Agent team | Several agents that work side by side as teammates, each with its own context, coordinated by the lead session. See Agent teams. |
| Agents view | The panel that lists your session, its subagents and background work. Open it with ← on an empty prompt or /tasks. See Agent view. |
| Approval dialog | The prompt that asks you to allow or refuse a tool call the rules do not settle. See Permission rules. |
| Auto memory | Notes Syntic Code keeps for itself about a project and reloads in later sessions. See Memory. |
| Auto mode | A permission mode where routine work runs without asking while high-impact actions still stop for you. See Permission modes. |
B
| Term | Meaning |
|---|---|
| Background session | A session started with --bg or /background that keeps running after you close the terminal. List them with syntic ps. See Background sessions. |
| Bare mode | --bare: a minimal start for scripts without settings hooks, plugins, skill discovery or auto memory. See CLI reference. |
C
| Term | Meaning |
|---|---|
| Channel | A way for a trusted MCP server to push a message into a running session. See Channels reference. |
| Checkpoint | A restore point taken at each of your messages. Rewind uses it to put files and the conversation back. See Checkpoints & rewind. |
| Compaction | Replacing older messages with a summary to free room in the context window. Happens automatically, or with /compact. See Context window. |
| Context window | How much text the model can consider at once: your messages, its replies, tool results, memory and instructions. See Context window. |
| Custom agent | A subagent you define in a Markdown file, with its own instructions, tools and model. See Subagents. |
D – F
| Term | Meaning |
|---|---|
| Deny rule | A permission rule that always refuses a matching tool call. Nothing, not even a hook, can override it. See Permission rules. |
| Effort | How hard the model thinks before answering: low up to max. Set it with /effort or --effort. See Models. |
| Fast mode | A switch for quicker replies on supported models. Toggle it with /fast or Alt+O. See Fast mode. |
| Folder trust | Your OK for a folder’s own settings, hooks and MCP servers to load. Asked the first time you open a folder. See Folder trust. |
G – H
| Term | Meaning |
|---|---|
| Goal | A condition set with /goal that keeps the agent working across turns until it is met. See Goals. |
| Headless run | A syntic -p run without the interactive screen, for scripts and CI. See Headless runs. |
| Hook | Your command, web request or model check that runs at a fixed point in a session, such as before a tool call. See Hooks and the Hooks reference. |
M
| Term | Meaning |
|---|---|
| Managed settings | Settings your organization deploys that you cannot override. See Managed settings. |
| Marketplace | A published list of plugins, with the source of each one. See Create a marketplace. |
| MCP | Model Context Protocol, the open standard for connecting outside tools and data. An MCP server adds tools, resources and prompts to your session. See MCP servers. |
| Memory | Instruction files, chiefly SYNTIC.md, loaded into every session. See Memory. |
| Mod | A small program that adds pieces to the interface, such as a status band or a panel, and reacts to session events. See Mods. |
O – P
| Term | Meaning |
|---|---|
| Output style | How replies are written: length, tone and format. See Output styles. |
| Permission mode | The overall stance on what runs without asking: default, acceptEdits, auto, plan, dontAsk or bypassPermissions. Cycle it with Shift+Tab. See Permission modes. |
| Plan mode | A permission mode where the agent explores and writes a plan but changes nothing until you approve. See Permission modes. |
| Plugin | A package of skills, commands, agents, hooks, MCP servers and more, installed in one step. See Plugins. |
| Prompt caching | Reusing the unchanged start of a request so later turns are faster and cheaper. See Prompt caching. |
R – S
| Term | Meaning |
|---|---|
| Restricted mode | --restricted: a locked-down session with no shell, code or web-fetch tools, and file tools kept to the current folder. See CLI reference. |
| Rewind | Going back to an earlier message, optionally restoring files. Esc Esc or /rewind. See Checkpoints & rewind. |
| Routine | A prompt that runs on a schedule for a project, managed with /schedule. See Scheduling. |
| Safe mode | --safe-mode: a troubleshooting start with memory files, plugins, skills, hooks and MCP servers turned off. See Debug your configuration. |
| Sandbox | An operating-system boundary that restricts which files the agent’s shell commands may touch and which hosts they may contact. See Shell sandbox. |
| Session | One saved conversation in one folder. You can resume, rename, branch or export it. See Sessions. |
| Settings scope | Which file a setting lives in: user (~/.syntic/settings.json), project (.syntic/settings.json), local (.syntic/settings.local.json) or managed policy. See Settings files. |
| Shell mode | Starting a prompt with ! to run a command yourself. See Shell mode. |
| Skill | A reusable set of instructions the agent applies when relevant, or that you run as /name. See Skills. |
| Slash command | A command typed at the prompt that starts with /, such as /model or /compact. See Slash commands. |
| Status line | A line under the prompt that runs your own script to show branch, cost or anything else. See Status line. |
| Subagent | A helper agent with its own context, started for one focused job. Its result comes back to the main conversation. See Subagents. |
SYNTIC.md | The project memory file. Committed to the repository so the whole team shares it. See Memory. |
T – W
| Term | Meaning |
|---|---|
| Tool | One action the agent can take, such as Read, Edit or Bash. See Tools. |
| Tool search | Loading tool descriptions only when needed, so many connected tools do not crowd the context window. See Tools. |
| Turn | One round of work: your message, then everything the agent does until it hands control back. |
| Workflow | A planned set of subagent steps that can run in parallel. See Workflows. |
| Worktree | A second checkout of a git repository on its own branch, so parallel work does not collide. See Worktrees. |