# Welcome to Pad > Pad is the multiplayer AI workspace. Every frontier model on one plan, chats you share when you choose, and pads that hold the documents, skills, and memory your team draws on. URL: https://pad.ai/docs These docs cover everything a member or an admin needs, from the first chat to organization settings. Start with the quickstart if you are new, or jump to the surface you have a question about. Sign in, start a chat, and invite your team in a few minutes. Organizations, pads, chats, groups, connections, skills, and memory. Talk to any model with your team's sources switched on. What a team knows: its Library, its skills, and its memory. Share access to the tools your team uses without sharing a key. Members, groups, security, billing, and usage. ## Guides [#guides] Step-by-step walkthroughs for the things most teams do first. Invite people, put them in groups, and give every group the pads and tools it needs. Turn the documents your team already has into knowledge every chat can draw on. Capture a task you do well as a playbook anyone can run with a slash. Connect a tool once and hand access to exactly the right people. Bring any tool with an MCP server into Pad, including services your own team runs. ## For agents [#for-agents] Every page here is available as Markdown: append `.md` to its URL, use the Copy page button, or read [llms.txt](/llms.txt) for an index and [llms-full.txt](/llms-full.txt) for the whole set. --- # Quickstart > Go from sign-in to a shared chat in a few minutes. URL: https://pad.ai/docs/quickstart ### Sign in [#sign-in] Open [app.pad.ai](https://app.pad.ai) and sign in. If you are the first person from your company on Pad, you create the organization as part of onboarding. If your company is already on Pad, ask an admin to invite you and follow the link in the invitation email. ### Add context [#add-context] Give Pad what your team works with. Under Connections, add the tools you use every day (a CRM, a calendar, a project tracker, or any MCP server) so chats can act in them. Then put the documents your team relies on into a pad's Library, where every chat that draws on the pad can search them. Skills and memory build up from there as you work. ### Invite your team [#invite-your-team] Invite people from Settings, put them in groups that mirror your org chart, and grant each group the pads and connections it needs. From then on, adding someone to a group hands them everything the group holds. ### Start a chat [#start-a-chat] Every chat starts with all of your sources on: the pads you belong to and the connections you hold. Pick a model, ask your question, and switch any source off from the chat header if you do not want it in the conversation. Bring a teammate in by mentioning them with ⌘ Enter, or share an in-organization link. ## Where to next [#where-to-next] * [Core concepts](/docs/concepts) explains the nouns you will meet everywhere in Pad. * [Chats](/docs/chats) covers sources, sharing, models, tools, and files. * [Set up your team](/docs/guides/set-up-your-team) walks an admin through the first week. --- # Core concepts > The nouns you will meet everywhere in Pad, and how they fit together. URL: https://pad.ai/docs/concepts | Concept | What it is | | ---------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Organization** | Your company's account: the tenant and billing boundary. Everyone signs in to one, and everything below belongs to it. | | **Pad** | A context container: one Library tree, its skills, and one memory pool, for a set of members. Every member has a personal pad of their own; teams share pads. | | **Chat** | A conversation with a model. Private to you until you share it, and scoped to the pads and connections you turn on as its sources. | | **Group** | A set of people that mirrors your org chart. Grant a group a pad or a connection and every member has it. | | **Connection** | A tool your organization connected once (a CRM, a calendar, a project tracker, an MCP server, or a website the chat's browser signs in to) and shared with the groups and people who need it. | | **Document** | A page, file, or folder in a pad's Library. | | **Skill** | A reusable playbook that teaches the model to do a task your way, run with a slash in any chat. | | **Memory** | Facts Pad learns from your chats, kept for you, for each pad, and for the organization. | | **Model** | One of the frontier models Pad offers, picked per chat. | ## How they fit together [#how-they-fit-together] A chat is where work happens. It reaches exactly the sources it names: the pads that lend it documents, skills, and memory, and the connections that lend it tools. Nothing runs inside a pad; a pad is what a team knows, and a chat is what a team does with it. Access follows the org chart. Sharing content means pad membership, sharing a tool means a connection grant, and sharing a conversation means an invitation or a link. Groups make the first two bulk operations, so adding someone to a group hands them every pad and tool the group holds. Privacy is the default. Your personal pad has exactly one member, your chats are yours until you invite someone, and a tool only ever acts under a named grant. When a chat is shared, its sources are always visible in the header, so a transcript can only ever contain content from sources every reader can see named. --- # Desktop app > Pad in a native window for macOS, Windows, and Linux. URL: https://pad.ai/docs/desktop The desktop app is the same product as [app.pad.ai](https://app.pad.ai) in a window of its own. It signs in with the same account, shows the same organizations, and keeps itself up to date quietly in the background. ## Install [#install] Download the build for your platform from the [Downloads page](/downloads). The app installs in seconds and opens to the sign-in page. ## What is different on desktop [#what-is-different-on-desktop] * **Tabs.** Open chats, documents, and pages side by side in one window. * **Find in page.** Press Cmd/Ctrl+F to search the page you are on, including a long chat transcript. * **Native menus and shortcuts.** The keyboard shortcuts on the [Search and shortcuts](/docs/search-and-shortcuts) page work the same way. --- # Set up your team > An admin's first week: invite people, put them in groups, and give every group what it needs. URL: https://pad.ai/docs/guides/set-up-your-team ### Invite everyone [#invite-everyone] From Settings and then Members, invite your team by email. Give the people who will run the organization the admin role. ### Create groups [#create-groups] Under Settings and then Groups, create a group per team that mirrors your org chart. Add people to the groups they belong to. ### Create pads for the teams that need them [#create-pads-for-the-teams-that-need-them] Every organization starts with a General pad that new members join automatically. Create a pad for each team with its own documents, skills, and memory, and grant the matching group a role in it. ### Connect the tools [#connect-the-tools] Add the tools your teams use from Connections and grant each one to the groups that need it. Nobody needs a key or a login of their own for shared accounts. From here, new hires need one action: add them to a group, and every pad and tool the group holds is theirs. --- # Build a team Library > Turn the documents your team already has into knowledge every chat can draw on. URL: https://pad.ai/docs/guides/build-a-team-library ### Pick the pad [#pick-the-pad] Open Library in the sidebar and switch to the team's pad. Its members are the people who will draw on what you add. ### Upload what exists [#upload-what-exists] Drag in the processes, brand guidelines, specs, and FAQs the team already keeps elsewhere. Pad extracts the text so chats can search them. ### Write the pages that were missing [#write-the-pages-that-were-missing] Create pages for what lived only in people's heads. Pages are collaborative, so the team can write them together. ### Let chats fill in the rest [#let-chats-fill-in-the-rest] When a chat produces something worth keeping, the model suggests a document for the pad. Editors publish the good ones and the Library grows from real work. --- # Write your first skill > Capture a task you do well as a playbook anyone can run with a slash. URL: https://pad.ai/docs/guides/write-your-first-skill ### Start from a task you repeat [#start-from-a-task-you-repeat] Pick something you already do well and often: writing a proposal, triaging a bug report, drafting release notes. ### Create the skill [#create-the-skill] Open Skills, switch to the pad the skill belongs in, and choose New. Name it, then write the instructions the way you would brief a new teammate: what good looks like, what to ask first, what to avoid. ### Add supporting files [#add-supporting-files] Attach templates, examples, or reference material the task needs. They travel with the skill. ### Run it [#run-it] In any chat that has the pad as a source, type a slash and pick the skill. Improve the instructions after each run; every save keeps a version. --- # Share a tool with a group > Connect a tool once and hand access to exactly the right people. URL: https://pad.ai/docs/guides/share-a-tool-with-a-group ### Connect the tool [#connect-the-tool] Open Connections and choose Add. Pick the tool from the catalog or paste an MCP server URL, then sign in once with the account the team should use. ### Grant the group [#grant-the-group] On the connection's Access page, add the group. Every member holds the tool from then on, and people added to the group later get it automatically. ### Use it in a chat [#use-it-in-a-chat] Group members find the connection on in their chats' sources. Actions that change something ask for approval, and every call records who asked and who approved. --- # Connect a custom MCP server > Bring any tool with an MCP server into Pad, from a vendor's official server to a service your own team runs. URL: https://pad.ai/docs/guides/connect-a-custom-mcp-server Any service that exposes a remote MCP server can become a connection: a vendor's official server, or one your team built for an internal system. Once connected, its tools work in chats like any connection from the catalog. ### Add the server [#add-the-server] Open Connections in the sidebar, choose Add connection, and pick Custom MCP server. Give it a name your team will recognize and paste the server URL, an https address such as `https://mcp.example.com/mcp`. Pad speaks both the Streamable HTTP and the SSE transports. ### Choose how it authenticates [#choose-how-it-authenticates] Pick the authentication the server expects: * **OAuth (automatic)** for servers that register clients on their own; you sign in on the server's page and Pad keeps the tokens. * **OAuth (your app's client)** when the vendor gave you a client id and secret to use. * **Bearer token** for a personal access token or an API key. * **Custom headers** for anything else the server needs on each request. * **None** for servers that take no credentials. ### Choose the account model [#choose-the-account-model] With a **Shared account**, everyone acts through the one account you connect. With **Personal accounts**, each member links their own before using the tools. Pick Personal accounts whenever what the tool does should be attributed to the person doing it. ### Connect and check the tools [#connect-and-check-the-tools] Finish by signing in or entering the credential. Pad reads the server's tool list into the connection's Tools tab, which shows what the model will be able to call and lets you turn individual tools off. ### Share it [#share-it] A new connection is private to you until you grant it. From its Access page, grant the groups and people who should hold it; it then appears as a source in their chats. ## Good to know [#good-to-know] * Actions that change something ask for approval before they run. See [Tools and approvals](/docs/chats/tools). * If the server's token expires or is revoked, the connection shows that it needs reauthorization and its creator is notified. Reconnect from the connection's page. * Credentials are never shown again after they are saved, not even to the person who entered them. --- # Chats > A chat is a conversation with a model that draws on your team's sources. It is yours until you share it. URL: https://pad.ai/docs/chats Chats live at the top of the sidebar. Each one belongs to you, not to a pad, and starts with every source you can reach switched on: the pads you belong to and the connections you hold. The chat header names those sources at all times. ## Starting a chat [#starting-a-chat] Press C, or choose Chats in the sidebar and start typing. Pick a model from the model picker, attach files if you have them, and send. Your first message creates the chat. ## What a chat can reach [#what-a-chat-can-reach] Every pad in scope lends the chat its documents, skills, and memory; every connection in scope lends it tools. The [Sources](/docs/chats/sources) page explains how to narrow that. When a task needs a website rather than an API, the model opens a [browser](/docs/chats/browser) of its own. ## While a turn runs [#while-a-turn-runs] You do not have to wait. Send a message while the model is working and it steers the turn: the current step finishes, anything the model had queued after it is dropped, and the model picks up your message right away. Press ⌘ Enter (Ctrl+Enter on Windows and Linux) to stop the turn and send instead, or press Stop to end it without saying anything. ## Working together [#working-together] Mention a teammate to bring them into the chat, or share an in-organization link. Everyone in a chat sees the same transcript and the same sources. See [Sharing a chat](/docs/chats/sharing). ## Finding chats again [#finding-chats-again] The Chats panel lists your chats grouped by the day of their last activity. Search (Cmd/Ctrl+K) finds a chat by title or content, and Favorites keep the ones you return to within reach. --- # Sources > The pads and connections a chat draws on, and how to switch them on and off. URL: https://pad.ai/docs/chats/sources Every chat starts with all of your sources on. The sources control in the chat header lists each one, and switching a source off takes it out of the conversation from that point on. ## What a source contributes [#what-a-source-contributes] * **A pad** lends its Library (searchable documents), its skills (run with a slash), and its memory. * **A connection** lends its tools, which the model can call during the conversation. ## Your personal pad [#your-personal-pad] Your personal pad stays in scope only while you are the only person in the chat. The first invitation drops it, documents, skills, and memory alike, so nothing private reaches a teammate by accident. The invite dialog says so before you confirm. ## Everyone sees the same sources [#everyone-sees-the-same-sources] When a chat is shared, its sources stay visible in the header for every participant. A transcript can only ever contain content from sources every reader can see named, which is why sources are never hidden. --- # Sharing a chat > Bring teammates into a chat with a mention or a link. URL: https://pad.ai/docs/chats/sharing A chat is private to you until you share it. There are two ways in, and both are limited to people in your organization. ## Mention and share [#mention-and-share] Type @ and pick a teammate. On its own, a mention only points at someone: they are named in the message, and if they are already in the chat they are notified. A plain mention never changes who can see the chat. To bring someone in with the mention, press ⌘ Enter (Ctrl+Enter on Windows and Linux) on their row in the picker instead of Enter, or click the Share pill the row shows. A "Will share" chip appears beside the composer, and sending the message invites them: they see the whole transcript from the start and can reply, add files, and add sources they hold. Remove the chip before sending to keep the mention without the invitation. ## Share a link [#share-a-link] The Share control in the chat header creates an in-organization link. Anyone in your organization who opens it joins the chat. Turn the link off to stop new people from joining; everyone already in the chat keeps their access. ## Who can share [#who-can-share] The chat's creator can always invite people. A participant can invite when they belong to one of the pads the chat draws on (or to any chat that draws on no pad). Only the creator turns the link on and off or removes participants. ## What changes when you share [#what-changes-when-you-share] Your personal pad leaves the chat's sources with the first invitation. Everything else stays as it was, and the header keeps naming the sources so every participant knows what the model can reach. --- # Models > Pick the right frontier model for each chat, or let Auto choose one per turn, and switch whenever you like. URL: https://pad.ai/docs/chats/models Pad offers the frontier models from the major labs side by side, on one plan. The model picker in the chat sets which one answers; you can change it at any point in a conversation. ## Choosing a model [#choosing-a-model] Draft with a fast model and reason with a stronger one in the same chat. The picker shows each model's name and lab; the [rate card](/docs/billing/rates) shows what each one costs in credits. ## Auto [#auto] When your organization has Pad Router on, the picker's first entry is **Auto**. Under Auto, Pad Router picks the model and the reasoning effort for every turn: a quick follow-up goes to a fast model, routine work to a standard one, and a hard question to a frontier model, at an effort to match. There is no effort control under Auto; the router chooses the effort for you. Auto has one dial, its optimization mode: * **Cost** spends the fewest tokens. * **Balanced** balances quality, speed, and cost. * **Intelligence** reaches for the most capable models on harder work. Pick a mode from the Auto entry's submenu. Your last pick is the mode your new chats start on, and new chats start on Auto unless you have starred another model as your default. Your admin chooses which modes are available, and which models Auto may use: see [Models](/docs/admin/models). Every reply Pad Router decided carries an **Auto** badge at the top showing the mode it was routed under. Your admin chooses whether to name the routed model: by default the badge does not, so answers are judged on their own merit; when the display is on, the badge names the model that answered and its effort in place of the mode. Replies from before you switched a chat to Auto keep their own model badges. Auto costs whatever the model it picked costs; there is no routing fee. ## The context gauge [#the-context-gauge] Every model has a context window. The gauge beside the composer shows how much of it the conversation is using, so you know when to start a new chat. Under Auto the gauge measures against a fixed window sized for routing rather than any one model's, so the router can keep the whole catalog available for most of a chat's life. --- # Tools and approvals > How the model uses the tools your connections lend a chat, and when it asks first. URL: https://pad.ai/docs/chats/tools A connection in a chat's sources lends the model that tool's actions: searching a tracker, reading a calendar, drafting an email. The transcript shows each call as it happens. ## Approvals [#approvals] Actions that change something outside Pad ask for approval before they run. The approval card appears inline in the chat and in your [Inbox](/docs/inbox). Approving runs the call; denying tells the model to find another way. Approvals go to the people who can vouch for the tool: for a connection, the person who brought it into the chat or anyone else who holds it. Every call records who requested it and who approved it. ## When the browser needs you [#when-the-browser-needs-you] Work in the [browser](/docs/chats/browser) has its own asks: a sign-in, a verification code, or a step the model should not take alone. They arrive the same way, as a card in the chat and a row in your Inbox, and the turn waits until you answer, take over, or skip. ## Linking your own account [#linking-your-own-account] Some connections act as each person's own account rather than a shared one. The first time such a tool is needed, the chat shows a card that links your account in a few clicks. See [Personal accounts](/docs/connections/personal-accounts). --- # Browser > Pad opens a cloud browser inside a chat and uses websites for you, signed in where you are. URL: https://pad.ai/docs/chats/browser When a task needs a website rather than an API, the model opens a browser of its own. It navigates, reads pages, including embedded chat widgets, clicks, types, fills forms, takes screenshots, and downloads files. Every step lands in the transcript as a compact row: the page as the model read it after each step sits behind the row, and screenshots open in the preview. ## The Browser pane [#the-browser-pane] The Browser button in the chat's top bar opens the live browser beside the transcript (full screen on a phone). It opens on its own the first time a turn uses the browser, and while the pane is closed the button's dot says what is happening: a steady dot for an open browser, a pulsing one while the agent is driving, a halo while a person is. ## Watching and taking over [#watching-and-taking-over] While the agent drives, you watch: the frame glows and ignores clicks so nobody fights the agent. Take over hands you the wheel, at once when the browser is idle or, during a turn, as soon as the current step finishes, and the turn waits for you. Hand back returns control, and so does sending a message, since the turn that starts is the agent's. Teammates in the chat see who is driving. ## When the browser needs you [#when-the-browser-needs-you] Sign-in walls, verification codes, and anything the model should not decide alone come to you: the chat shows a card and your [Inbox](/docs/inbox) gets a row. Sign in inside the live browser and press Done, answer a one-time code or pick a sign-in method right in the card, or press Skip to let the model try another way. When the site is one of your website connections, the sign-in is kept for next time. The browser has no security key and no passkey, and cannot use yours: it runs in the cloud with nothing plugged in and no phone nearby. A site that asks for one is steered to its other methods, an authenticator code (which Pad can type from a saved authenticator key), a prompt to your phone, a text, or an email, and the model asks you when the site offers none. If an account signs in only with a security key, add an authenticator app to it before connecting. ## Signed-in sites [#signed-in-sites] Connect a website once and the browser starts every chat already signed in, for every website connection you hold among the chat's sources. See [Website connections](/docs/connections/websites). ## Sessions, recordings, and cost [#sessions-recordings-and-cost] A chat has one browser at a time. It closes after ten minutes idle and reopens when needed; End session in the pane's menu closes it now. Sessions are recorded, a replay plus the step screenshots, unless a connection in use is marked sensitive. Each browser step costs credits at the rate the [rate card](/docs/billing/rates) shows. Code the model runs in the chat's sandbox can drive the same browser, so a many-step task can run as a script while you watch. --- # Files and attachments > Attach files to a chat, preview them, and find everything a chat has shared. URL: https://pad.ai/docs/chats/files Drop a file onto the composer or use the attach button to add it to a message. The model reads it as part of the conversation. ## The Files panel [#the-files-panel] Everything attached in a chat is listed in its Files panel, with previews for documents, images, spreadsheets, and code. Files attached to a chat are private to its participants. ## Documents from the Library [#documents-from-the-library] Mention a document with @ to bring a page from a pad's Library into the conversation without attaching a copy. The chat links to the live document, so edits made later are reflected. --- # Pads > A pad is what a team knows: its Library, its skills, and its memory, for the people who share it. URL: https://pad.ai/docs/pads A pad holds exactly three things: a Library tree of documents, a set of skills, and a memory pool. Its members are people and groups. Chats draw on pads as sources; nothing runs inside one. ## Personal and shared pads [#personal-and-shared-pads] Every member has a personal pad with exactly one member: you. It is where private documents, skills, and memory live, and it can never be shared. Shared pads belong to teams. They are invite-only, or open to everyone in the organization so anyone can join. Organizations start with a shared pad called General that new members join automatically. ## Switching pads [#switching-pads] Library, Skills, and Memory in the sidebar share one active pad. The switcher at the top of their panels changes it for all three, and each device remembers the pad you last picked. ## Creating a pad [#creating-a-pad] Use New pad in the switcher or Settings and then Pads. Organization admins can always create shared pads; members can when the organization allows it. --- # Members > Who is in a pad, how they got there, and what each role can do. URL: https://pad.ai/docs/pads/members A pad's members are people and groups. People can be added directly; groups bring every one of their members along, and those rows show the group they came from. ## Pad roles [#pad-roles] | Role | What it adds | | ---------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Member** | Browse and search published documents, draft documents and publish their own, create and edit skills, and use the pad as a source in chats. | | **Editor** | Everything a member can do, plus publish or unpublish any document (including drafts the model suggested), curate the pad's memory, and manage skill listings. | | **Admin** | Everything an editor can do, plus manage membership and settings, and archive the pad. | A person who reaches a pad through more than one route holds the highest role among them. ## Managing members [#managing-members] Pad admins add and remove members and change roles from the pad's Members tab in Settings. Group-sourced rows are managed from the group itself (Settings and then Groups). --- # Library > The documents a pad holds: pages, files, and folders, edited together and searched by every chat that draws on the pad. URL: https://pad.ai/docs/pads/library The Library is a pad's tree of documents. Pages are rich text you write in Pad, files are uploads, and folders keep them organized. ## Writing together [#writing-together] Pages are collaborative: several people can edit the same page at once and see each other's changes as they type. Every page keeps a history of its revisions. ## Uploading files [#uploading-files] Drag files into the tree or use Upload. Pad extracts the text so chats can search them like any other document. ## Suggested documents [#suggested-documents] The model can draft a document from a chat, for example a summary or a process it just helped write. Drafts land in the pad as suggestions for an editor to publish, edit, or discard, so the Library stays trustworthy. ## Searching and linking [#searching-and-linking] Search (Cmd/Ctrl+K) finds documents across every pad you belong to. In a chat, mention a document with @ to bring it into the conversation; in a page, link to other pages the same way. --- # Skills > Reusable playbooks that teach the model to do a task your way. URL: https://pad.ai/docs/pads/skills A skill is a named, versioned package of instructions, with any files the task needs alongside. Skills live in pads and follow the open agent-skills conventions, so they import and export as plain folders. ## Using a skill [#using-a-skill] Type a slash in the composer and pick the skill. The model follows its instructions for that message, drawing on the pad's documents and the chat's tools as the skill directs. ## Creating a skill [#creating-a-skill] Open Skills in the sidebar, pick the pad, and choose New. Write the instructions, add supporting files, and save. Every save creates a new version, so a skill can be improved without losing what worked before. ## Importing skills [#importing-skills] Choose Import on the Skills surface to bring skills in from outside Pad: * **Upload** a zip or tar.gz of skill folders, or a single SKILL.md file. * **GitHub**: paste the URL of a public repository, optionally pointing at a branch or a folder inside it, such as `github.com/owner/repo/tree/main/skills`. Pad lists every skill it finds in the source. Pick the ones you want and import them. A skill whose name already exists in the pad is imported as that skill's next version rather than as a duplicate. ## Discover [#discover] The Discover surface lists built-in skills and the skills other pads in your organization have listed, ready to add to the pad you are in. ## Exporting [#exporting] Every skill exports as a plain folder archive, SKILL.md and its files, so it travels to any other tool that follows the same conventions. --- # Memory > What Pad learns from your chats, where it goes, and how to take anything back. URL: https://pad.ai/docs/pads/memory Memory is the set of facts Pad learns as you work: preferences, decisions, context about your team. It fills in on its own after a chat settles; nothing waits on you. ## Three pools [#three-pools] * **Mine** follows you: what Pad learns about you and your preferences. * **Pads** hold what a team draws on: facts learned in chats that use the pad. * **Organization** holds what applies to everyone. The Memory surface in the sidebar shows each pool, with a tab per kind. ## Seeing what was learned [#seeing-what-was-learned] Under a settled turn, a Learned chip lists the facts the chat produced. Each one can be kept, retracted, or moved to your own memory. ## The disclosure window [#the-disclosure-window] When a chat's participants are fewer than a pad's members, a fact learned for that pad stays visible only to the chat for a short window before it discloses to the whole pad. Anyone in the chat can retract it during the window. Organization admins set the window's length in Settings. ## Curating memory [#curating-memory] Facts are versioned and never deleted for good. Edit, forget, and restore them from the Memory surface. Pad admins choose in the pad's settings whether chats may write into its pool. --- # Connections > The tools your organization connected once and shares with the groups and people who need them. URL: https://pad.ai/docs/connections A connection is a configured integration owned by your organization: a CRM, a calendar, a project tracker, a database, any tool with an MCP server, or a website the chat's browser signs in to. It is granted to the organization, to groups, or to individual people, and whoever holds it can use its tools in the chats they bring it into. Nobody ever sees a key. ## Where connections live [#where-connections-live] Connections has its own row in the sidebar. The list shows every connection in the organization; each one has a General page, an Access page for grants, and a Tools page listing what it lends a chat. ## Websites [#websites] A [website connection](/docs/connections/websites) keeps a signed-in session for a site with no API, so the chat's [browser](/docs/chats/browser) can work inside it as you. ## Shared and personal accounts [#shared-and-personal-accounts] Some tools run through one shared account the organization set up. Others act as each person's own account, which you link once, the first time you need it. See [Personal accounts](/docs/connections/personal-accounts). --- # Connect a tool > Add a connection from the catalog or from an MCP server URL. URL: https://pad.ai/docs/connections/connect ### Choose the tool [#choose-the-tool] Open Connections in the sidebar and choose Add connection. Pick the tool from the catalog, or choose Custom MCP server and paste its URL for anything else, including services your own team built (the [custom MCP guide](/docs/guides/connect-a-custom-mcp-server) walks through it). For a site the model should use through its browser, choose Connect a website instead; see [Website connections](/docs/connections/websites). A site that signs in through Google, Microsoft, or another provider takes **Single sign-on** as its sign-in method there, with the connection it signs in with picked or created in the same step. ### Sign in once [#sign-in-once] For a shared account, sign in to the tool as the account the whole team should use. For tools that act as each person, you sign in with your own account and each teammate links theirs later. ### Share it [#share-it] Grant the connection to the groups and people who need it from its Access page. See [Share access](/docs/connections/access). ## Who can manage a connection [#who-can-manage-a-connection] The person who created a connection manages its settings and grants. Once it is granted beyond its creator, organization admins can manage it too, and only admins can grant a connection to the whole organization. --- # Website connections > Sign in to a website once, and the chat's browser starts signed in every time it uses the site. URL: https://pad.ai/docs/connections/websites Some tools have no API, or the work only happens in a browser. A website connection keeps a signed-in session for a site so the model can work inside it without asking you to log in each time. Pad holds the session, and any saved login, encrypted; the browser provider never does. ## Connecting a website [#connecting-a-website] From Connections choose Add connection, then Connect a website. Enter the site's address and, if it differs, its sign-in page. Website connections use personal accounts by default, so each person signs in as themselves; choose a shared account for a team login a whole group should use. ## Signing in [#signing-in] Open the connection's Browser tab and choose Sign in. A real browser opens on the site's sign-in page. Sign in as you normally would, two-factor prompts included, and press Done. Pad keeps the signed-in session, not what you typed, and nothing from a sign-in is ever recorded. The same flow works from a link: a single-use link, valid for fifteen minutes, that you can open on your phone. Sign in there, press Done, and your other devices update on their own. ## Saving your login [#saving-your-login] After signing in, Pad offers to save your password, and an authenticator key if the site asks for one-time codes, so the browser can sign in again without you when the session expires. This is optional. A saved login is typed into the site's own sign-in page by Pad and is never shown to the model or to anyone. The authenticator key is the setup key the site shows when you turn on its authenticator app (the text offered beside the QR code, a long string of letters and digits) or the otpauth link behind that code; Pad computes each one-time code from it. In a password manager it is the value of the item's one-time password field when you edit the item. The six-digit code itself expires in seconds and is not the key, and 1Password's secret reference (an op\:// address) is not either; the form says so if you paste one. ## Signing in with Google, Microsoft, or another connection [#signing-in-with-google-microsoft-or-another-connection] Many sites sign in through an identity provider instead of a password of their own. When you connect such a site, choose **Single sign-on** as its sign-in method and pick the connection it signs in with, or create one on the spot from the Google, Microsoft, GitHub, Apple, or Okta presets. The sign-in connection is an ordinary website connection: sign in to it once on its Browser tab, save its login there if you like, and every site linked to it uses that sign-in from then on. Update the login in one place and every linked site follows. A linked site keeps no saved login of its own. The browser opens the site already signed in to the provider, so the site's "Sign in with Google" button is one click, and the model can press it and fill the provider's saved login when a session expires. The link is a choice of the people who manage the site; both connections must use the same account mode, and a site can use another site as its sign-in connection only when you have access to both. A sign-in connection's page lists the sites that use it. Archiving it leaves those sites without a sign-in method until a manager chooses another, and the archive confirmation names them. ## What chats get [#what-chats-get] A chat's browser starts signed in to every website connection you hold among the chat's sources. A signed-in session is yours alone: a teammate's turn in the same chat uses their own sessions, never yours. ## Staying signed in [#staying-signed-in] The Browser tab shows whether you are signed in, signed out, or whether a saved login was rejected. Sign in again from the same tab when a site signs you out. When the model signs in with your saved login during a chat, including through an authenticator code or a sign-in connection, that session is kept the same way. A session unused for 45 days is dropped. Sign out from the tab at any time; leaving the organization or archiving the connection removes every session and saved login. ## Settings for a website [#settings-for-a-website] On the connection's General tab: **Sensitive** turns off recordings and screenshots for the site and limits watching to the person driving. **Approve actions** asks before every click or submit on the site. **Enhanced stealth** uses extra measures on sites that block automated browsers. --- # Share access > Grant a connection to the organization, to groups, or to people. URL: https://pad.ai/docs/connections/access A connection's Access page lists who holds it. Add a grant to hand a tool to a group or a person; remove one to take it back. ## Grants [#grants] * **The organization.** Everyone holds the connection. Organization admins only. * **A group.** Every member of the group holds it, now and as the group changes. * **A person.** One individual holds it. ## What holding a connection means [#what-holding-a-connection-means] A holder can use the connection's tools in any chat that has it as a source, bring it into chats they take part in, and approve its tool calls. A chat keeps a connection as a source only while the person who brought it in still holds it. --- # Personal accounts > Tools that act as you, with an account you link once. URL: https://pad.ai/docs/connections/personal-accounts Some tools should act as each person rather than as a shared account: email, calendars, and anything where who did it matters. For those connections, each holder links their own account. ## Linking your account [#linking-your-account] Open the connection from Connections and choose Link account, or wait until a chat needs the tool: the chat shows a card that links your account in a few clicks, then carries on. ## What the model can do [#what-the-model-can-do] With a personal account linked, the tool acts only as you, in chats you take part in. Unlink the account from the connection's page at any time. --- # Inbox > One place for everything that needs you: approvals, suggested documents, memory notices, and mentions. URL: https://pad.ai/docs/inbox The Inbox at the top of the sidebar collects the rows that would otherwise get lost in chats. ## What lands here [#what-lands-here] * **Tool approvals** waiting on you. * **Browser asks**: a sign-in, a one-time code, or a choice the [browser](/docs/chats/browser) needs from you. * **Suggested documents** the model drafted for a pad you edit. * **Memory notices** about what Pad learned, for your information only. * **Mentions** of you in chats. * **Runs that stopped** because a usage limit was reached. The unread count shows on the Inbox row. Act on a row from the Inbox, or open the chat or document it came from. --- # Search and shortcuts > The command palette and the keyboard shortcuts that move you around Pad. URL: https://pad.ai/docs/search-and-shortcuts ## Search [#search] Press Cmd/Ctrl+K anywhere to open the command palette. It searches documents, chats, and skills you have access to, and it also navigates: type a surface or a setting to jump to it, or Connect to add a tool. ## Keyboard shortcuts [#keyboard-shortcuts] | Shortcut | What it does | | ---------------------------------- | ---------------------------------------- | | Cmd/Ctrl+K | Open search | | Enter (while a turn runs) | Steer the running turn with your message | | Cmd/Ctrl+Enter (while a turn runs) | Stop the turn and send your message | | C | Create a chat or a document | | Cmd/Ctrl+B | Show or hide the sidebar | | Cmd/Ctrl+F | Find in the current page | | @ | Mention a person or a document in a chat | | / | Run a skill in a chat | --- # Organization settings > What owners and admins manage, and where to find it. URL: https://pad.ai/docs/admin Settings, reached from the organization switcher at the top of the sidebar, holds everything that shapes the organization. ## The settings pages [#the-settings-pages] | Page | What it holds | | ------- | ------------------------------------------------------------------------------------------------------------------------------------ | | General | The organization's name and URL, and the memory disclosure window. | | Models | Pad Router: whether Auto is offered, the modes members may pick, whether replies name the routed model, and the models Auto may use. | | Members | Everyone in the organization, their roles, and pending invitations. | | Groups | The sets of people that grant pads and connections in bulk. | | Pads | Every shared pad, with its settings and members. | | Billing | The plan, credits, and payment details. | | Usage | Credits used across the organization. | ## Organization roles [#organization-roles] | Role | What it adds | | ---------- | ------------------------------------------------------------------------------------------------------------------------------------------- | | **Member** | Use Pad: chats, the pads they belong to, the connections they hold, and open pads they choose to join. | | **Admin** | Everything a member can do, plus manage members, groups, and invitations, every shared pad and connection, and the organization's settings. | | **Owner** | Everything an admin can do, plus billing and ownership of the organization itself. | Every organization has at least one owner. --- # Models and Pad Router > Turn Pad Router on or off, choose the modes members may pick, decide whether replies name the routed model, and set which models Auto may use. URL: https://pad.ai/docs/admin/models Settings and then Models holds the organization's Pad Router settings. Owners and admins change them; members can read them. A change applies to the next turn of every chat on Auto. ## Enable Pad Router [#enable-pad-router] While Pad Router is on, the model picker's first entry is [Auto](/docs/chats/models#auto): Pad Router picks the model and the reasoning effort for every turn, and new chats start on Auto unless the member has starred another model as their default. Turning it off removes Auto from the picker. A chat already on Auto keeps working on Pad's default model until someone picks another, and its earlier replies keep their Auto badges. ## Optimization modes [#optimization-modes] Auto's one dial. Members pick from the modes you leave on: | Mode | What it optimizes | | ---------------- | --------------------------------------------------- | | **Cost** | Spends the fewest tokens. | | **Balanced** | Balances quality, speed, and cost. | | **Intelligence** | Reaches for the most capable models on harder work. | At least one mode stays on. A chat on a mode you turned off routes under Balanced, or under the first mode left on when Balanced is off too, until someone picks another. ## Show routed model [#show-routed-model] Off by default, and recommended. Every reply Pad Router decided carries an Auto badge showing the mode it was routed under, so answers are judged on their own merit rather than by model name. Turn it on and the badge names the model that answered and its effort in place of the mode. Naming is a display choice, not a secret: the Usage page reports credits by model either way. ## Models Auto may use [#models-auto-may-use] The list is every model Pad Router can route to, grouped by lab and marked with its tier (frontier, standard, or fast). Uncheck a model and Auto never picks it; members can still pick it by hand from the model picker. At least one model stays checked. Leaving many models unchecked degrades routing: Pad Router chooses among what is left, so a mode may not reach the tier it was built for. Uncheck a model for a reason (a lab your policy excludes, a model you are still evaluating), not by habit. --- # Members and roles > Invite people, set their organization role, and remove them when they leave. URL: https://pad.ai/docs/admin/members ## Inviting people [#inviting-people] From Settings and then Members, invite by email. Each invitation carries an organization role and arrives as a link from [no-reply@notifications.pad.ai](mailto:no-reply@notifications.pad.ai); an invited person who signs in with that email joins the organization. If an invitation does not arrive, see [Security and sign-in](/docs/admin/security#if-a-sign-in-email-does-not-arrive). ## Changing roles [#changing-roles] Owners and admins change roles from the same page. See [Organization settings](/docs/admin) for what each role can do. ## Removing someone [#removing-someone] Removing a member signs them out everywhere and ends their memberships and grants. Their contributions stay: documents, skills, and chat messages remain where they are. --- # Groups > Sets of people that mirror your org chart and grant pads and connections in bulk. URL: https://pad.ai/docs/admin/groups A group is an organization-level set of people, such as Design, Sales, or Platform. Granting a group a role in a pad gives every member that role; granting a group a connection makes every member a holder. Add someone to a group and every pad and tool the group holds is theirs. ## Managing groups [#managing-groups] Owners and admins create groups and edit their member lists from Settings and then Groups. Each group has a General tab, a Members tab, and a Pads tab listing the pads it holds and the role it holds them with. ## Groups in pads [#groups-in-pads] A pad admin grants a group a role from the pad's Members tab. Rows that came through a group show the group's name and are managed from the group, not the pad. --- # Managing pads > Create shared pads, set how people join them, and archive the ones a team no longer needs. URL: https://pad.ai/docs/admin/pads Settings and then Pads lists every shared pad in the organization. Organization admins are implicitly admins of every shared pad. ## Creating a pad [#creating-a-pad] Choose New pad, name it, and pick an icon. Organization admins can always create shared pads; whether members can is an organization setting. ## Join policy [#join-policy] A pad is either invite-only or open to the organization. Open pads are visible to everyone, who can join on their own. The General pad adds every new member automatically. ## Pad settings [#pad-settings] A pad's settings page holds its name and icon, its join policy, and its memory write policy, which decides whether chats may write into the pad's memory pool. Archiving a pad hides it and its contents without deleting them. --- # Security and sign-in > How people sign in to Pad, and how to manage where you are signed in. URL: https://pad.ai/docs/admin/security ## Sign-in [#sign-in] Pad's sign-in is hosted: you sign in on a Pad-branded page run by our identity provider, and Pad never stores a password. Sign in with your Google account or with your email address and a code. Single sign-on and multi-factor authentication are available to organizations that need them; with single sign-on, people sign in with their work account and no email is involved. ## If a sign-in email does not arrive [#if-a-sign-in-email-does-not-arrive] Verification codes, invitations, and password resets come from [no-reply@notifications.pad.ai](mailto:no-reply@notifications.pad.ai). If one does not arrive, check your spam or quarantine folder first. On Microsoft 365, a message can sit in quarantine until your IT team releases it or allows the sender, so ask them to allow notifications.pad.ai. Signing in with Google sends no email, and organizations on single sign-on never wait on one. ## Your sessions [#your-sessions] Settings and then Security lists every device where you are signed in. Sign out of any one of them, or out of every device at once. ## Access control [#access-control] Access in Pad follows memberships and grants, never per-file permissions: pad membership for content, connection grants for tools, invitations and links for chats. See [Core concepts](/docs/concepts). --- # Plans and credits > One per-seat plan with every model included, paid for in credits. URL: https://pad.ai/docs/billing Pad has one plan. Each seat comes with a monthly allowance of credits, which every model call spends according to the [rate card](/docs/billing/rates). New organizations start with a free trial and a starting balance of credits. Current prices are on the [pricing page](/pricing). ## Credits [#credits] Credits are shared by the organization. When the monthly allowance runs low, buy a credit pack, or turn on auto-reload so a pack is added whenever the balance drops below a threshold you set. ## Managing billing [#managing-billing] Organization owners manage the subscription, payment method, and invoices from Settings and then Billing. --- # Usage > See where the organization's credits go. URL: https://pad.ai/docs/billing/usage Settings and then Usage charts credits spent over time, by member and by model. Owners and admins see the whole organization; members see their own usage. --- # Model rates > What each model costs in credits. URL: https://pad.ai/docs/billing/rates Every model has a published rate per token, for input, output, and cached input. The rate card in Settings lists the current rate of every model Pad offers, and anyone in the organization can view it. Rates follow the providers' own prices and can change when a provider's do; the card always shows what applies today. --- # FAQ > Short answers to the questions teams ask first. URL: https://pad.ai/docs/faq ## Which models can I use? [#which-models-can-i-use] The frontier models from the major labs, side by side, on one plan. Pick one per chat and switch whenever you like, or leave the picker on [Auto](/docs/chats/models#auto) and let Pad Router choose the model and the effort for each turn. ## Can we turn Auto off, or keep it away from some models? [#can-we-turn-auto-off-or-keep-it-away-from-some-models] Yes. Owners and admins manage Pad Router under Settings and then Models: turn it off, choose which optimization modes members may pick, decide whether replies name the routed model, and uncheck any model Auto may not use. See [Models](/docs/admin/models). ## Is my data used to train models? [#is-my-data-used-to-train-models] No. Pad calls the providers' APIs under terms that exclude training, and your documents, chats, and memory stay in formats no single lab owns. ## Can teammates see my chats? [#can-teammates-see-my-chats] Not until you share them. A chat is private to you until you mention someone in it or share an in-organization link, and your personal pad leaves the chat's sources with the first invitation. ## What if a tool I use is not in the catalog? [#what-if-a-tool-i-use-is-not-in-the-catalog] If it has an MCP server, paste the URL and Pad connects to it: see [Connect a custom MCP server](/docs/guides/connect-a-custom-mcp-server). The catalog grows from what teams request. ## Can Pad use a tool that has no API? [#can-pad-use-a-tool-that-has-no-api] Yes. The model opens a [browser](/docs/chats/browser) of its own and works on the site the way you would. Connect the site as a [website connection](/docs/connections/websites) so the browser starts signed in. ## Can it sign in with Google? [#can-it-sign-in-with-google] Yes. Give the site **Single sign-on** as its sign-in method and pick the Google connection it signs in with (or create one from the preset). Sign in to Google once on that connection's Browser tab; every site linked to it then signs in through it, and the model can press the site's "Sign in with Google" button itself. See [Website connections](/docs/connections/websites). ## I did not get the sign-in email [#i-did-not-get-the-sign-in-email] Check spam and quarantine first: sign-in codes and invitations come from [no-reply@notifications.pad.ai](mailto:no-reply@notifications.pad.ai), and a Microsoft 365 filter can hold them until your IT team releases the message or allows the sender. Signing in with Google sends no email, and organizations on single sign-on never wait on one. See [Security and sign-in](/docs/admin/security#if-a-sign-in-email-does-not-arrive). ## How do I get help? [#how-do-i-get-help] See [Support](/docs/support). --- # Glossary > Short definitions of the words Pad uses. URL: https://pad.ai/docs/glossary | Term | Meaning | | ---------------------- | ----------------------------------------------------------------------------------------------------------------------------- | | **Approval** | A confirmation the model waits for before a tool call that changes something. | | **Auto** | The model picker's entry that lets Pad Router choose the model and the effort for each turn. | | **Browser** | The cloud browser a chat opens when a task needs a website; you can watch it, take over, and hand it back. | | **Chat** | A conversation with a model, private until shared. | | **Connection** | A tool the organization connected and shares through grants. | | **Credits** | The unit every model call spends; each seat comes with a monthly allowance. | | **Disclosure window** | The period in which a fact learned for a pad stays visible only to the chat that produced it. | | **Document** | A page, file, or folder in a pad's Library. | | **Grant** | The act of giving the organization, a group, or a person a connection. | | **Group** | An organization-level set of people that grants pads and connections in bulk. | | **Holder** | Someone a connection has been granted to, directly or through a group. | | **Library** | A pad's tree of documents. | | **MCP server** | A service that exposes tools over the Model Context Protocol; any one can become a connection. | | **Memory** | Facts Pad learns from chats, kept per person, per pad, and per organization. | | **Optimization mode** | Auto's one dial: Cost, Balanced, or Intelligence. | | **Organization** | Your company's account and billing boundary. | | **Pad** | A context container: one Library, its skills, and one memory pool, for a set of members. | | **Pad Router** | The part of Pad that picks a model and a reasoning effort per turn when a chat is on Auto. | | **Personal pad** | The one pad that belongs to a single person and can never be shared. | | **Sign-in connection** | A website connection another website connection signs in with, such as Google for a site with a "Sign in with Google" button. | | **Skill** | A versioned playbook run with a slash in a chat. | | **Source** | A pad or connection a chat draws on. | | **Steer** | A message sent while a turn runs; the model takes it up at its next step. | | **Take over** | Taking the wheel of a chat's browser from the agent; Hand back returns it. | | **Website connection** | A connection that keeps a signed-in session for a site, for the chat's browser. | --- # Support > How to reach us. URL: https://pad.ai/docs/support * **Message us in the app.** The messenger in the corner of every Pad page reaches the team directly. * **Email.** Write to [hello@pad.ai](mailto:hello@pad.ai). * **Book a demo.** Talk to us about rolling Pad out to your organization from the [site](/).