What it is

A toggle in Customize → Memory, with two settings: "search and reference chats" and "generate memory from chats."

Why it matters

Without it, every new chat starts from zero context. With it on, Claude recalls prior conversations so you don't have to re-explain background each time.

How it works

Once both toggles are on, Claude automatically references relevant past chats at the start of a new one — likened to catching up with a friend you haven't seen in months, rather than re-introducing yourself.

• Depending on your Claude version you may see one or two toggle switches under Memory — either is fine as long as both are enabled.

What it is

A dedicated workspace for a recurring task, with its own instructions, uploaded reference files and memory — separate from your regular chat history.

Why it matters

For any repeatable output (same format, different content each time), Projects let you set instructions once instead of re-explaining context every chat — saving tokens and improving consistency over time.

How it works (demoed live)

1. Create a new Project and give it a short description of its purpose.
2. Go to Instructions and write out the task in detail: role, format, structure, tone, constraints.
3. Upload supporting reference material to Context (e.g. past scripts) so Claude learns the writer's voice — any file type works.
4. From then on, only new inputs are needed; the Project retains all standing instructions and reference material.

• Instructions can be granular — e.g. "only take the writing style from this document, not the content."
• Ask Claude to output section by section (intro → items → outro) so you can approve each part before moving on.
• Projects aren't transferable — they stay tied to your account, unlike Skills.
• Not limited to writing — usable for competitor analysis, review dashboards, or any recurring analytical task.

What it is

An in-chat deep-research feature that searches and cross-references many web sources before answering, rather than a quick single-pass response.

Why it matters

Useful for well-sourced, thorough analysis — market sizing, competitor research, sentiment — at the cost of much higher token usage and longer wait time.

1. In a new chat, click "+" and toggle on Research (separate from the lighter, faster "web search").
2. Write a detailed prompt with role, task and context — always assign Claude a role.
3. Approve any requested connector permissions if prompted.
4. Claude opens a "Lead Researcher" panel showing its live plan and source-gathering progress — often 400+ sources.
5. Output includes a TL;DR, market segmentation, growth/size figures, white-space analysis, and a full clickable source list.

• May not be available on the free plan, due to its higher context/token cost.
• Recommended cadence: occasional — "once or twice a month at most," best at the start of a new campaign, product or IP.
• Cites and cross-references real sources, so it's presented as less prone to hallucination than a standard reply.

What it is

Turns Claude's output into an actual interactive app, site or game rather than plain text — "Claude coding a website or app for you," akin to vibe coding.

Why it matters

Reports and research become interactive and demo-able — for a client meeting, say — instead of a static document or slide deck.

• Top prompting tip (works for any feature): add "If you feel like asking me more questions that would help you produce a better output, please ask me." Claimed to improve output quality ~20–30%.
• Also useful for internal/educational tools — e.g. an interactive RAG-concept explainer built for students.

What it is

A reusable, portable instruction file (markdown) Claude loads only when relevant, invoked by typing "/" in chat.

Why it matters

Conceptually similar to Projects — both are custom instructions — but a Skill is transferable: downloadable and shareable with anyone, where a Project stays local to your account.

1. Go to Customize → Browse to see default skills (Canvas Design, Learn, Brand Guidelines, MCP Builder, Theme Factory, …).
2. To build a custom skill, use the built-in Skill Creator: describe the goal, optionally attach reference documents, and Claude generates a skill file through guided Q&A.
3. Skills can be downloaded as .md files and re-uploaded elsewhere, or found online (community GitHub repos) and imported via Customize → Skills → Add.

• Anthropic also ships domain skill packs via Customize → Browse → Plugins — e.g. a Marketing pack ("Brand Review," "Campaign Plan," "Competitive Brief") activated in one click, then invoked with "/".
• A skill file is just an instructions document — arbitrarily narrow ("write UGC ad scripts") or broad (a teaching methodology).

What it is

The mechanism that lets Claude take real-world actions in external tools, rather than only processing information. "Claude is the brain, connectors are its hands and legs."

Why it matters

Without connectors Claude can only reason and generate text — it can't book something, edit a file in another app, or update a calendar. Connectors close that gap.

1. Customize → Connectors to browse pre-built integrations (Drive, Gmail, Calendar, Canva, Microsoft 365, Slack, Figma, …) and click Connect to authorize.
2. For tools outside the directory: Customize → Connectors → Add Custom Connector, paste the tool's MCP server URL, name it, add it.
3. Once connected, the tool's actions are available directly in chat or a Skill — e.g. resizing an uploaded image via a connected Canva skill ("/canva, resize to 9:16 and 3:4").

• "Connectors" and "MCPs" are used somewhat interchangeably — connectors = what's in Claude's default directory; MCP = the underlying open protocol, used for a custom/third-party integration.
• Most MCPs are free — you pay only for the token usage of the conversation, not for the integration itself.
• Custom skills can layer on top of custom connectors to standardize prompt formats for a specific tool.

What it is

The desktop version of Claude, combining memory, Projects, Artifacts, Research Mode, Skills and Connectors in one environment — plus local file access and scheduled, recurring autonomous tasks.

Why it matters

Where Claude moves from a single-response tool to an "AI employee" — given a goal, not just a task, and left to execute it, including on a schedule, without the chat window staying open.

Local file access

1. Switch to Cowork mode and click Add New Folder to grant access to a directory (Downloads, Desktop, …).
2. Claude can then read, cross-reference and act on files within that folder directly.

Scheduled, recurring tasks ("loops")

1. Go to Scheduled → New Task; set it up manually or let Claude configure it conversationally.

Demoed live — calendar + notes → follow-up action

1. Prerequisite: Google Calendar already connected via Connectors.
2. Prompt: "Go to my calendar, today, execute all action items on my meeting with [X]."
3. Cowork located the relevant meeting-notes document (from an AI notetaker), extracted action items, and autonomously scheduled a real follow-up meeting and drafted a related document (e.g. a term-sheet-style MOU) — without further step-by-step instruction.

• Requires a paid Claude plan — not available on the free tier.
• A related feature, Dispatch, lets you queue a task from the mobile app while away from your desktop, ready by the time you're back — mentioned only briefly as worth exploring independently.