TUE, JUNE 16, 2026
Independent · In‑Depth · Practitioner‑Tested
✎ General

Grok Agent Settings Explained 2026 — Every Customisation Option and What It Actually Does

Grok has four customisation layers: Custom Instructions (global, 4,000 chars), Custom Agents (4 slots, persona-level, March 2026), Grok Skills (task-level /commands, May 2026), and Workspaces (context isolation per project). Explains every setting location, what each layer is for, what to put where, and the settings most users never find — Fun Mode, Heavy mode toggle, and workspace instruction overrides.

By AIToolsRecap June 16, 2026 7 min read 10 views
Home Articles General Grok Agent Settings Explained 2026: Every Optio...
Grok Agent Settings Explained 2026 — Every Customisation Option and What It Actually Does

ALL SETTINGS LOCATIONS

Custom Agents: Profile icon → Settings → Customize → Create/Edit Agents
Grok Skills: Settings → Skills → Create New Skill
Custom Instructions: Settings → Customize → Custom Instructions
Workspaces: Main navigation sidebar → Your Workspaces
Model selection: Chat interface → model dropdown (SuperGrok shows all models)
Memory: Settings → Privacy → Memory

The Full Grok Customisation System — How the Pieces Fit Together

Grok has four distinct customisation layers that work at different levels of scope. Understanding which to use for which purpose saves a lot of wasted configuration:

Layer Scope Best for Limit
Custom Instructions All conversations globally Your identity, persistent context ("I am a developer in Vietnam") ~4,000 chars
Custom Agents Agent-specific conversations Named specialist personas with distinct roles and behaviours 4 slots, 4,000 chars each
Grok Skills Task-level persistent instructions Repeatable workflows invoked with /command Multiple skills, ~4,000 chars each
Workspaces Context isolation per project Separating work, personal, side project contexts with different files and history Multiple workspaces

Custom Instructions — The Global Layer

Custom Instructions is a single global block that applies to every Grok conversation, regardless of which agent or workspace you are in. Think of it as the persistent context that follows you everywhere: your role, your preferences, things Grok should always or never do for you specifically.

The instruction limit was reduced from 12,000 characters to ~4,000 characters when Custom Agents launched in March 2026. This was intentional — Custom Instructions was never meant to be a place to paste detailed task instructions (that belongs in Skills). Use Custom Instructions for:

DO put here: Your role ("I am a founder of a B2B SaaS startup"), your location and language preferences, persistent context that saves you re-explaining yourself (your company name, target audience, tech stack), universal format preferences ("always output in English even if I ask in Vietnamese"), and global negative rules ("never add a conclusion unless I ask").

DON'T put here: Task-specific instructions (use Skills), agent persona details (use Custom Agents), project-specific context (use Workspaces). These dilute the global instruction effectiveness and waste the 4,000-character limit.

Custom Agents — The Persona Layer

Custom Agents (launched March 4, 2026) let you create up to 4 named AI personas, each with its own 4,000-character instruction set. Each agent is a specialist — a named entity with its own role, tone, and behavioural rules. You select which agent to use at the start of a conversation, or invoke by name mid-conversation.

Setting What it does Best practice
Name How you invoke the agent in conversation Short, specific name. "ATLAS" not "My Research Helper"
Description Helps Grok's multi-agent system route to this agent by context One specific sentence: "Use when asked to research topics with web and X sources"
Instructions The system prompt for this agent (4,000 char max) Use the ALWAYS/NEVER/FORMAT/TONE structure. Negative rules prevent drift most effectively
Auto-engage Whether the agent activates by context without being called by name Enable for your most-used agent. Disable for specialised agents to prevent wrong activations

Grok Skills — The Task Layer

Grok Skills (launched May 18, 2026) are persistent instruction bundles for specific repeatable tasks — invoked with a /command slash in conversation. A Skill is different from an Agent: an Agent is a full persona with a consistent identity across a conversation; a Skill is a task specification you invoke on demand within any conversation with any agent.

Skill vs Agent decision rule: If you need a specialist persona for an ongoing conversation (a research partner, a code reviewer, a writing editor) → use an Agent. If you need to execute a specific repeatable workflow on demand within any conversation (produce a weekly report, format an email, run a competitive brief) → use a Skill.

Settings path: Profile icon → Settings → Skills → Create New Skill. Three fields: Name (the slash command, e.g. /monitor), Description (when to activate), Instructions (the task specification, up to ~4,000 characters). See our full Skills template guide for 8 complete ready-to-use Skills.

Workspaces — The Context Isolation Layer

Workspaces provide full context isolation — separate file uploads, separate conversation history, and separate instruction overrides for each workspace. A Work Workspace can contain your company style guide and technical glossary; a Personal Workspace can contain different reference documents entirely. Workspace-level instructions override your global Custom Instructions for conversations within that workspace.

Grokipedia confirms Workspaces take context isolation a step further than Claude Projects by fully separating files and conversation history within each workspace. For users who want strict separation between work, personal, and project contexts — without files from one bleeding into another — Workspaces is the mechanism.

Model Settings — What Changes with SuperGrok

Setting Free X Premium+ ($40/mo) SuperGrok ($30/mo)
Model access Grok 4.3 (limited) All models All models + priority
Custom Agents Basic 4 slots 4 slots + potential for more
Image generation Limited Yes Unlimited
Aurora video gen No No Yes
Rate limits Low Medium High
Heavy mode (4-agent debate) Limited Yes Yes + priority

Settings Most Users Never Find

Fun Mode / Spicy Mode

Settings → Customize → Personality. Enables Grok's unfiltered, deliberately irreverent mode. Relaxes some safety guardrails. The most Grok-specific setting — no equivalent exists in Claude or ChatGPT.

Memory toggle

Settings → Privacy → Memory. Controls whether Grok remembers context across conversations. Disable if you want each conversation to start fresh with no carryover. Enable for a persistent working relationship where Grok remembers your preferences and history.

Heavy mode manual toggle

In the chat interface, toggle Heavy mode to force Grok to activate its full 4-agent parallel reasoning architecture for your next response. Slower but produces substantially better analysis on complex questions. Only available on paid plans.

Workspace-level instruction override

Inside any Workspace, you can set workspace-specific instructions that override the global Custom Instructions for conversations in that workspace. Found in Workspace settings → Customize. Use to set a completely different context per project without touching your global settings.

Setup guide: How to Create Custom Grok Agents · Agent library: 50 ready-to-use Grok agents · Skills templates: Grok Skills instruction templates

Tags
GrokAI agentsBest AI ToolsAI GuideGenerative AI2026

Spot an inaccuracy?

We verify facts before publishing and correct errors promptly. If something in this article is wrong or outdated, let us know.

Report an error →