Grok

One conversation for questions, writing, coding, images and files.

Best for

An everyday task that does not need live Search, Imagine or Multi-agent.

Real example

Upload a PDF and ask for its argument, weak evidence and questions to raise.

Start with

Explain the document's argument and separate what it proves from what it merely claims.

Where you make it better

Check quotations against the original document.

Searches the live web and can incorporate real-time public posts from X.

Best for

Breaking information, current events and public conversation.

Real example

Compare an official product recall with customer reports appearing on X.

Start with

Use official sources for facts and X only to characterise reported experience or reaction.

Where you make it better

Live does not mean reliable; inspect sources and manipulation risk.

Runs several separate AI researchers and combines their work with source links.

Best for

A large question that can be divided into independent searches.

Real example

Compare a costly system across technical fit, price, users and supplier stability.

Start with

Give each researcher a different question and add one final reviewer whose job is to challenge the combined report.

Where you make it better

Check whether every researcher started with the same mistaken assumption.

Generates and edits images and videos from text or reference media.

Best for

Concept art, storyboards, visual experiments and short motion studies.

Real example

Turn a packaging sketch into three photographed shelf concepts and a short rotation video.

Start with

Preserve the product proportions while changing only [specific elements].

Where you make it better

Check rights, likeness, text accuracy and whether the media could mislead.

Low-latency natural conversation with optional live camera in supported apps.

Best for

Hands-free thinking, rehearsal and discussing what you can see.

Real example

Practise a sales call while Grok plays a sceptical prospect.

Start with

Stay in character, ask one question at a time and give feedback after the role-play.

Where you make it better

Review any factual detail that was spoken quickly or inferred visually.

Analyses PDFs, documents, screenshots, photos and diagrams.

Best for

Grounding a question in material you supply.

Real example

Compare a site plan, progress photos and a written issue list.

Start with

Describe observable differences only. Mark anything that requires an on-site inspection.

Where you make it better

Do not use image interpretation as a substitute for qualified inspection.

A longer-form writing surface with inline editing and suggestions.

Best for

Developing a substantial piece of writing beyond a scrolling chat.

Real example

Shape an article through outline, argument review and paragraph-level revision.

Start with

Challenge the outline and evidence before polishing the prose.

Where you make it better

Keep the point of view and final editorial choices human.

Remembers preferences and context across chats when enabled.

Best for

Avoiding repeated personal preferences and working conventions.

Real example

Remember preferred units, dietary constraints and briefing format.

Start with

Remember only these stable preferences: [list]. Do not infer new personal facts.

Where you make it better

Inspect, correct and delete memory that is wrong or too personal.

An early-beta coding agent and CLI with plan review, diffs, plugins and sub-agents.

Best for

Professional software engineering in an existing repository.

Real example

Find a latency regression using parallel code, infrastructure and database investigations.

Start with

Use plan mode. Reproduce the issue and show evidence before proposing edits.

Where you make it better

Approve the plan, review every diff and control tool permissions.

Lets developers add Grok search, vision, voice, image or video features to their own app.

Best for

A product that needs one Grok capability with its own controls around it.

Real example

Add current, source-linked market information to an internal analyst tool.

Start with

Return the same fields every time, include source links and mark uncertainty clearly.

Where you make it better

Add validation, privacy rules, spending limits and approval before the app acts.