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