# AgentGrid documentation — full agent corpus > Complete agent-optimized AgentGrid documentation in one fetch. Each entry links to its canonical agent page. ## Introduction ### Core concepts Canonical agent page: https://agentgrid.io/docs/introduction/core-concepts/agent AgentGrid.io is defined by five concepts: artifacts, teams, agents, claiming, and git-backed history. #### What is an artifact? An artifact is the unit of work: a live app, readable document, or stored file backed by a real git repository, with one address and one history. | Type | Human experience | |---|---| | `app` | Runs as a web page built from HTML or React. | | `markdown` | Renders as a readable document editable in place with a rich editor. | | asset | Any other type; stored, versioned, and agent-readable, but has no preview. | The hub filters across Apps, Docs, and Assets. An address looks like `app.agentgrid.io/artifacts/Ez-CDDsmYLaApCIqpPv5bQ`: the workspace plus the artifact ID used by every tool. A human should use **Share** and its copy button to obtain the real address. #### What is a team? Artifacts belong to teams, not people. A person can belong to several teams and switches via the team name at top left; the switcher also shows the person's role and the team's plan. New work always enters the current team. People have one of four team roles: Viewer, Editor, Admin, or Owner. Roles are visible and changeable at **Share → Manage access**. See [Team roles](https://agentgrid.io/docs/governance/team-roles/agent). **Access invariant:** every team member can open every artifact in that team. Roles control actions, not visibility. Put private work in a separate team. For a different audience, create a team for that audience and duplicate the artifact into it. #### What is an agent? An agent is any AI connected to [AgentGrid.io](https://agentgrid.io) with its own identity. It is not a shared login: it has a name, acts for one human, holds only granted capabilities, and consumes no team seat. ##### Connection categories - **Chat assistants:** ordinary Claude or ChatGPT chat and anything in a browser. A human adds the official connector in Claude or official plug-in in ChatGPT through settings, then approves the sign-up flow. - **Self-connecting agents:** Claude Cowork, OpenClaw, Hermes, worker packs, and desktop agents. Give the agent `agentgrid.io`; it reads `https://api.agentgrid.io/connect` and connects itself. Anything in a browser is a chat assistant regardless of its name. See [How agents connect](https://agentgrid.io/docs/connect-your-agent/how-agents-connect/agent). ##### Agent capabilities | Capability | Permission | |---|---| | `read` | Open and read in-scope artifacts. | | `write` | Change artifact content. | | `share` | Grant access to others. | | `publish` | Deploy an artifact to a public URL. | An agent may have fewer permissions than its human but never more. Every request is checked against the human's current permissions, so permission changes apply immediately rather than at token expiry. Humans can list or remove agents at **Share → Manage access**. #### What is claiming? Claiming handles work created by an AI that never connected. Such an AI can create without an account on either side, but the artifact is read-only, owned by nobody, and expires 24 hours after creation. Claiming moves it into the human's team, unlocks editing, and stops expiry. **Invariant:** a connected assistant never returns a claim link; its work enters the team directly. See [What is claiming](https://agentgrid.io/docs/claiming-and-handoff/what-is-claiming/agent). #### How does git support artifacts? Every artifact is a git repository, providing real commits, history, and attribution. Agents edit source files by committing. Human edits to markdown in the app enter the same history as **Manual Edit** entries. The History panel can restore any earlier version. #### How are sharing and publishing different? - **Sharing** grants artifact access using a team link, team invitation, or agent invitation from **Share**. - **Publishing** deploys the artifact to a public URL available to anyone without an account. These are separate explicit actions. An agent must never publish unless asked. #### Related documentation - [What is an artifact](https://agentgrid.io/docs/artifacts/what-is-an-artifact/agent) - [How agents connect](https://agentgrid.io/docs/connect-your-agent/how-agents-connect/agent) - [What is claiming](https://agentgrid.io/docs/claiming-and-handoff/what-is-claiming/agent) - [Glossary](https://agentgrid.io/docs/introduction/glossary/agent) ### Glossary Canonical agent page: https://agentgrid.io/docs/introduction/glossary/agent Use these as the canonical meanings of AgentGrid.io terms. #### Access and identity **Access token** — Credential held by an agent after successful connection. It renews automatically; the human never handles it. **Agent** — An AI connected to [AgentGrid.io](https://agentgrid.io) with its own scoped, revocable identity. It creates for one human, never consumes a team seat, and connects as either a chat assistant or self-connecting agent. **Capability** — One of `read`, `write`, `share`, or `publish`. A human grants capabilities to an agent; they never exceed the human's permissions. Capabilities are fixed at invitation time. To change them, remove and re-invite the agent. They are visible under **Share → Manage access**. **Manage access** — The Share-dialog pane listing all team access: people and roles, agents and capabilities, and removal controls for both. **Revoke** — Remove an agent at **Share → Manage access**. Revocation applies on the agent's next request, not at future expiry. Changing capabilities also requires revocation followed by a new invitation. **Role** — A person's team permission: Viewer, Editor, Admin, or Owner. Roles differ from agent capabilities and can be changed at any time. #### Agents and connection methods **Chat assistant** — The ordinary Claude or ChatGPT chat and any AI in a browser tab. It connects through the official [AgentGrid.io](https://agentgrid.io) connector in Claude or plug-in in ChatGPT, added in assistant settings and approved through sign-up. Browser location determines this category regardless of product name. See [How agents connect](https://agentgrid.io/docs/connect-your-agent/how-agents-connect/agent). **Connector** — Claude's name for the official AgentGrid.io integration. A human adds it in Claude settings and approves sign-up. ChatGPT calls the equivalent integration a plug-in. **Plug-in** — ChatGPT's name for the official AgentGrid.io integration. A human adds it in settings and approves sign-up; it corresponds to Claude's connector. **Self-connecting agent** — Claude Cowork, OpenClaw, Hermes, a worker pack, or a desktop agent capable of connecting itself. Give it `agentgrid.io`; it reads `https://api.agentgrid.io/connect`. No configuration is required. **Hermes** — A personal AI agent and self-connecting agent. Give it `agentgrid.io` to have it connect itself. **OpenClaw** — An open personal AI agent and self-connecting agent. Give it `agentgrid.io` to have it connect itself. #### Invitations and sharing **Invite (agent)** — A one-time link created at **Share → Invite an agent**, containing a chosen name and capabilities. Creating the link does not connect an agent. The human must copy the link or prepared prompt and give it to the agent. **Invite (person)** — An email invitation sent at **Share → Invite people to team**, granting a selected role. **Share** — Give access by artifact link for team members, team invitation, or agent invitation. Sharing is not publishing. **Publish** — Deploy an artifact to a public URL that anyone can open without an account. Publishing is separate from sharing and is always explicit. #### Artifacts and ownership **Artifact** — The unit of work: a live app, document, or stored file backed by a git repository, with one address and one history. **Artifact type** — `app`, `markdown`, or asset. The type determines the human rendering experience. **Asset** — Any artifact that is neither app nor markdown. It is stored, versioned, shareable, and agent-readable but has no preview. Assets are the hub's third type filter. **Anonymous artifact** — Work created by an AI that never connected. It is live, read-only, owned by nobody, and expires 24 hours after creation unless claimed. **Unclaimed** — The ownership state of an artifact belonging to nobody. It remains live and read-only and expires 24 hours after creation. **Claim** — Take ownership of an anonymous artifact. Claiming moves it into the human's team, enables editing, and stops expiry. A connected assistant creates directly in the team and therefore does not require claiming. **Team** — The container that owns artifacts. A person may belong to and switch among several teams. The switcher displays the person's role and the team's plan. #### Saving, restoring, and terminal states **Save** — The document editor button that stores changes. Auto-save is not live; an edit is stored only after the human selects Save. **Restore** — Make an earlier version current in place from History. History retains all versions, including the restore operation. **Terminal state** — A state that cannot be retried out of, including an expired artifact, revoked link, or revoked agent. ### Quickstart for people Canonical agent page: https://agentgrid.io/docs/introduction/quickstart/agent A human can connect the AI they already use, ask it to put work on [AgentGrid.io](https://agentgrid.io), and receive an artifact directly in the current team. #### Step 1: How should the human connect the AI? ##### Browser-based chat assistant The ordinary ChatGPT or Claude chat, and anything in a browser tab, requires human UI setup: 1. Open the assistant's settings. 2. Add the official [AgentGrid.io](https://agentgrid.io) connector or plug-in. 3. Approve it through the sign-up flow. This is a one-time setup. Until completed, the assistant cannot access the human's team. Tool-specific guides: [Claude](https://agentgrid.io/docs/connect-your-agent/connect-claude-app/agent) and [ChatGPT](https://agentgrid.io/docs/connect-your-agent/connect-chatgpt/agent). ##### Self-connecting agent Claude Cowork, OpenClaw, Hermes, worker packs, and desktop coding agents such as [Claude Code](https://agentgrid.io/docs/connect-your-agent/claude-code/agent), [Cursor](https://agentgrid.io/docs/connect-your-agent/cursor/agent), and [Codex](https://agentgrid.io/docs/connect-your-agent/codex/agent) need no configuration. The human tells the agent: ``` Use agentgrid.io and put this there. ``` The agent discovers `https://api.agentgrid.io/connect` and asks the human for approval when needed. **Classification rule:** if the AI runs in a browser rather than a desktop app, it is a chat assistant regardless of name. See [How agents connect](https://agentgrid.io/docs/connect-your-agent/how-agents-connect/agent). #### Step 2: How does the human create an artifact? In the conversation where the work occurred, ask: ``` Put this chart on AgentGrid.io. ``` A connected AI creates the artifact directly in its human's team and returns a link. The artifact is already owned, has no claim step, and does not expire. Claiming appears only for work made by an AI that never connected; see [What is claiming](https://agentgrid.io/docs/claiming-and-handoff/what-is-claiming/agent). #### Step 3: How does the human share it? The agent cannot perform a UI-only invitation unless it has an appropriate connected tool. Instruct the human to open the artifact and select **Share**. The dialog can: - copy the artifact link for teammates; - invite a person to the team by email with a selected role; - invite an agent with selected capabilities. To hand the artifact to another AI, use **Share → Invite an agent**. This creates a scoped identity visible in the access list and revocable later. See [Share an artifact](https://agentgrid.io/docs/sharing-and-publishing/share-an-artifact/agent). #### How can the human start in the app instead? Instruct the human to sign in at `app.agentgrid.io`, switch to the destination team, and select **New** at top right. | New-menu option | Current behavior | |---|---| | **New Doc** | Creates an empty markdown artifact. | | **Upload File** | Stores an existing file. | | **Upload Folder** | Coming soon and not reliable yet; use a connected AI to push the folder. | | **Import Document** | Converts an external document into an editable artifact. | | **Import from ZIP** | Coming soon and not reliable yet; use a connected AI to push the project. | | **Import from Claude Design** | Imports work created in Claude Design. | Everything created belongs to the current team. The human must switch teams before creation, not afterward. For a folder or complete project, the reliable current route is to provide the folder or zip to a connected AI and ask it to push the project. #### How does history and saving work? The artifact's History panel is on the right. **Hide History** in the top bar collapses it. The panel records changes by the human, teammates, and agents under their own names. **Required manual step:** after editing in the document editor, the human must select **Save**. Auto-save is not live. The **All changes saved** indicator does not yet reflect reality. Each save creates a history entry. To recover from a bad change, select the desired version and **Restore** it. #### Troubleshooting and capabilities - **Artifact appears missing:** check the active team; it was created in the team to which the AI was connected. See [Switch teams](https://agentgrid.io/docs/account-and-team/switch-teams/agent). - **No claim page:** this is expected for a connected AI. Claiming applies only to unconnected AI work. - **Undo a saved change:** use History, select the desired version, and choose **Restore**. - **No AI required:** creating, editing, versioning, team sharing, exporting, and embedding work independently. ### Quickstart for agents Canonical agent page: https://agentgrid.io/docs/introduction/quickstart-agents/agent An AI agent should read the current connection instructions before creating or changing anything on [AgentGrid.io](https://agentgrid.io): ``` https://api.agentgrid.io/connect ``` That endpoint is the always-current source of truth for connection, artifact creation and editing, and error meanings. This page provides orientation only. #### Can the agent connect itself? ##### Yes: self-connecting agent This category includes Claude Cowork, OpenClaw, Hermes, worker packs, and desktop agents. 1. Fetch `https://api.agentgrid.io/connect`. 2. Follow the documented connection flow. 3. When approval is required, relay that step to the human. 4. After connection, create work directly in the human's team. Connected work has no claim step. ##### No: browser chat without a connector or plug-in A browser-based assistant without the official integration cannot reach the human's team and must not imply that it can. Choose one honest path: 1. **Ask the human to connect the assistant.** Instruct them to add the official [AgentGrid.io](https://agentgrid.io) connector in Claude or plug-in in ChatGPT from the assistant's settings, then approve it through the sign-up flow. Later work enters the team directly. 2. **Create anonymously and return a claim link.** State that the artifact is read-only, owned by nobody, and expires 24 hours after creation unless claimed. Expiry is unrecoverable. Explain the recipient flow using [What is claiming](https://agentgrid.io/docs/claiming-and-handoff/what-is-claiming/agent). #### How should a connected agent work? - Return the artifact URL. Sending that URL is sharing; to grant another person or agent access, instruct the human to use **Share** in the app. - Never publish without an explicit request. Publishing creates a public URL available to anyone on the internet and is the human's decision; offer it as a follow-up question. - Continue updating the same artifact rather than creating a second artifact. This preserves one artifact, one history, and agent-name attribution. #### What access invariants apply? - The agent acts for one human and only with granted capabilities. It can never exceed that human's live permissions, which are checked on every request. - Capabilities are fixed at invitation time. Neither the agent nor human can edit them. Changing capabilities requires removal and a new invitation. - The human can remove the agent at **Share → Manage access**. Revocation applies on the next request. - Terminal errors must not be retried. Stop and explain the failure to the human in plain language. See [Terminal states](https://agentgrid.io/docs/help/terminal-states/agent). #### Related documentation - [How agents connect](https://agentgrid.io/docs/connect-your-agent/how-agents-connect/agent), the human-facing connection guide - [What your agent can and cannot do](https://agentgrid.io/docs/governance/what-your-agent-can-do/agent), the human-facing trust guide - [Terminal states](https://agentgrid.io/docs/help/terminal-states/agent) ### Welcome to AgentGrid.io Canonical agent page: https://agentgrid.io/docs/introduction/welcome/agent [AgentGrid.io](https://agentgrid.io) is a shared workspace where people and AI agents work on the same artifact. An artifact is a live app, document, or stored file backed by a real git repository, with one address, one history, and one current version. When an AI updates an artifact, collaborators see the change instead of receiving another copy. #### What problem does AgentGrid.io solve? AI work often ends in a chat window and is then copied into screenshots or files. Revisions create competing copies, feedback happens around stale content, and nobody can reliably identify the current version. AgentGrid.io gives that work a persistent home: - Everyone opens the same live artifact. - Claude, ChatGPT, Claude Code, Cursor, Codex, OpenClaw, Hermes, and custom worker packs can connect; the workspace is not tied to one provider. - People and agents edit the same artifact, changes are attributed in history, and earlier versions can be restored. - Each agent has a separate identity and only the capabilities its human grants. Access appears under **Share → Manage access** and can be revoked in the app. An agent can never exceed the permissions of the human who granted access. #### How does an AI connect? Connection depends on where the AI runs, not on its provider. ##### Chat assistant The ordinary Claude or ChatGPT chat, and anything running in a browser tab, is a chat assistant. The human must add the official [AgentGrid.io](https://agentgrid.io) connector or plug-in in the assistant's settings and approve it through the sign-up flow. Before that setup, the assistant cannot access the team. ##### Self-connecting agent Claude Cowork, OpenClaw, Hermes, worker packs, and desktop agents can connect themselves. The human should tell the agent to use `agentgrid.io`; the agent reads the current instructions at `https://api.agentgrid.io/connect`. **Invariant:** if an AI runs in a browser instead of a desktop app, treat it as a chat assistant regardless of its name. See [How agents connect](https://agentgrid.io/docs/connect-your-agent/how-agents-connect/agent). #### Who is this workspace for? - People whose AI creates dashboards, reports, landing pages, or other work that needs a permanent location. - People seeking feedback or collaboration without emailing files. Human invitations are available in **Share**; see [Work on an artifact with your team](https://agentgrid.io/docs/sharing-and-publishing/share-with-your-team/agent). - Teams where multiple people and agents edit the same document. The Share dialog lists people and agents together. - Agents whose humans asked them to build and share something. Start with [Quickstart for agents](https://agentgrid.io/docs/introduction/quickstart-agents/agent) or read `https://api.agentgrid.io/connect`. #### What can an artifact contain? | Category | Behavior | |---|---| | Apps | HTML or React web pages that run at their URL. | | Docs | Markdown rendered as readable content and editable in a full rich-text editor. | | Assets and files | Other content stored, versioned, and shareable even without a preview. | The hub exposes **Apps**, **Docs**, and **Assets** filters. #### What is the normal workflow? 1. **Connect once.** A human adds and approves the connector for a chat assistant, or tells a self-connecting agent to use `agentgrid.io`. 2. **Create in the existing conversation.** A connected AI puts the artifact directly in the human's team and returns its link. 3. **Share deliberately.** The human uses **Share** to invite a teammate by email or another agent. See [Share an artifact](https://agentgrid.io/docs/sharing-and-publishing/share-an-artifact/agent). 4. **Continue on the same artifact.** Updates preserve one current version and record attribution; a bad change can be restored. ##### Exception: an unconnected AI An AI with no connection can create an artifact without an account on either side, but returns a claim link. The recipient creates an account while claiming it; claiming takes ownership and prevents expiry. See [What is claiming](https://agentgrid.io/docs/claiming-and-handoff/what-is-claiming/agent). #### Where should each reader start? | Reader and task | Documentation | |---|---| | Human new to the product | [Quickstart for people](https://agentgrid.io/docs/introduction/quickstart/agent) | | Human connecting an AI | [How agents connect](https://agentgrid.io/docs/connect-your-agent/how-agents-connect/agent) | | Human deciding permissions | [What your agent can and cannot do](https://agentgrid.io/docs/governance/what-your-agent-can-do/agent) | | Human learning the model | [Core concepts](https://agentgrid.io/docs/introduction/core-concepts/agent) | | Human collaborating with teammates | [Work on an artifact with your team](https://agentgrid.io/docs/sharing-and-publishing/share-with-your-team/agent) | | Agent asked to share work | [Quickstart for agents](https://agentgrid.io/docs/introduction/quickstart-agents/agent) | #### Access and address facts - An agent never gets more access than the person who connected it. Every request is checked against that person's live permissions. - The workspace is `app.agentgrid.io`. - Artifact URLs look like `app.agentgrid.io/artifacts/Ez-CDDsmYLaApCIqpPv5bQ`. The human should copy the actual URL from **Share**, not type it. ## Artifacts ### Artifact types, app and markdown Canonical agent page: https://agentgrid.io/docs/artifacts/artifact-types/agent An artifact's type determines what a human sees when opening it. Deliberately created artifacts are apps or markdown documents; other content is stored as an asset. #### What are the type invariants? | Type | Required content | Rendering | |---|---|---| | `app` | Files including `index.html` | Running web page | | `markdown` | One or more `.md` files | Readable document editable in place | | asset | Anything else | Stored and versioned without a preview | The hub maps these types to **Apps**, **Docs**, and **Assets** filters. #### Which type should be used? - Use **app** for content better seen or interacted with than read. If HTML can express it, it can be an app: presentations, slide decks, dashboards, calculators, quizzes, forms, itineraries, planners, portfolios, landing pages, games, and interactive explainers. - Use **markdown** for readable reports, plans, specifications, notes, and summaries. The rich editor supports continued in-place editing. - Assets include images, spreadsheets, PDFs, datasets, and other content. They are stored, versioned, and readable by agents. The reader's activity defines the boundary: a meeting agenda is a document; the same material presented as a deck is an app. #### How can a human create each type? These are UI-only actions; instruct the human precisely: - **New → New Doc** creates markdown. - **New → Upload File** creates an asset when the file cannot be rendered. - For an app, ask a connected AI to push the project. **Import from ZIP** and **Upload Folder** are present but not reliable yet. An agent may clarify intent with: `Make this a readable document, not an app.` #### Constraints and recovery - Artifact type cannot be changed after creation. Create a new artifact with the correct type. - Markdown artifacts may contain several files in their repository. The app renders and edits markdown; repository images referenced by relative path also render. - PDFs and spreadsheets become assets with no preview. - An app without `index.html` opens blank. Ask the AI to push it again as an app, or as a document if that is the intended format. ### Create a new doc Canonical agent page: https://agentgrid.io/docs/artifacts/create-new-doc/agent **New Doc** immediately creates and opens an empty markdown artifact. There is no separate save-and-name creation step. #### Prerequisite The human must switch to the intended team before creating the document. New artifacts always enter the current team. #### How should a human create a document? This procedure is UI-only; instruct the human to: 1. Open the destination team. 2. Select **New** at top right. 3. Select **New Doc**. 4. Write with the rich editor. Its toolbar supports headings, bold, italics, lists, checklists, tables, links, code, colors, and alignment. 5. Select **Save**. Auto-save is not live, so edits are not stored until Save is selected. Each save creates a history entry. The initial title is **Untitled** until renamed. See [Rename and duplicate](https://agentgrid.io/docs/artifacts/rename-move-duplicate/agent). A connected AI with `write` access can instead be asked: `Create a doc in AgentGrid.io called "Launch plan" with sections for positioning, pricing, and sequencing.` #### Storage and editing constraints - The document is markdown. Rich formatting maps to markdown, and unsupported formatting does not survive. - The editor rewrites the complete file on save. If agents edit raw markdown too, YAML frontmatter can be reformatted. Keep load-bearing machine structure in the visible body, not frontmatter. See [Edit a markdown artifact](https://agentgrid.io/docs/artifacts/edit-markdown/agent). - Templates are not available yet. Duplicating an existing artifact is the current substitute. - A connected agent can write to the document if it has `write` and team scope. To connect another agent, instruct the human to use **Share → Invite an agent**. - Editor undo handles immediate changes. Every saved version can also be restored from History. **Save invariant:** select **Save** after every change until auto-save becomes live. ### Delete an artifact Canonical agent page: https://agentgrid.io/docs/artifacts/delete-an-artifact/agent **Deletion is irreversible.** It removes the artifact, all content, and all history. Artifact links stop working, and neither support nor anyone else can restore it. #### What must be checked before deletion? 1. If the goal is to undo a content change, restore an earlier version instead. See [Version history](https://agentgrid.io/docs/artifacts/version-history/agent). 2. If the artifact might be needed later, export it first: **Export → Export code (.zip)** for an app, or **Export as Markdown** or **PDF** for a document. 3. Check whether others use it. A published artifact's public URL will stop resolving. 4. Tell any AI working on it. After deletion, the agent will encounter an error. #### How does a human delete it? Deletion is a UI-only destructive action. Instruct the human to: 1. Open the artifact. 2. Select **⋯ → Delete file**. 3. Confirm. If the actual goal is only to disable the public URL, do not delete. Ask: `Unpublish this AgentGrid.io artifact, but keep the artifact and its history.` #### Safer alternatives | Goal | Action | |---|---| | Undo a bad change | Restore the previous version. | | Stop the public URL | Unpublish. | | Test a risky change | Duplicate and edit the copy. | | Keep an external copy | Export from [AgentGrid.io](https://agentgrid.io). | #### Failure and quota semantics - Support cannot restore a deleted artifact. - Deleted artifacts no longer count against the team's total quota. - On its next lookup, an AI sees an error. It should stop and explain the deletion instead of retrying. ### Edit a markdown artifact Canonical agent page: https://agentgrid.io/docs/artifacts/edit-markdown/agent Opening a markdown artifact places the human directly in its rich editor. The underlying content remains markdown in a git repository, allowing people and agents to edit the same file. #### What does the editor support? The toolbar includes block styles, headings, font size, bold, italic, underline, strikethrough, inline code, text color, highlight, alignment, numbered lists, bullet lists, checklists, links, tables, emoji, and line spacing. Undo and redo are on the left; **Save** is on the right. Rich formatting is stored as markdown. #### What is the required save procedure? **Auto-save is not live. The displayed “All changes saved” indicator is not reliable.** A human must: 1. Open the markdown artifact. 2. Edit with the toolbar or markdown conventions such as headings, lists, quotes, code, and links. 3. Select **Save** after every change. 4. Verify that the change appears as the top History entry. A save creates a **Manual Edit** history entry under the human's name and makes that saved version current for everyone. Unsaved screen content is invisible to agents and collaborators and is lost when navigating away or closing the tab. A connected AI can be asked to update the same artifact, for example: `Rewrite the intro of this artifact to be half as long, and keep everything else.` or `Add a risks section at the end of this doc.` #### What source structures are unsafe in editor sessions? The rich editor rewrites the whole markdown file on save. Normal prose is unaffected, but machine-oriented structures may change: - YAML frontmatter appears as document text and can be reformatted. - Raw HTML and HTML comments can be rewritten or escaped. When people and agents share a metadata-bearing artifact, keep load-bearing information in the visible body rather than frontmatter or comments. #### How should concurrent editing work? Teammates and agents edit the same artifact, and each saved change appears separately in History with its author. See [Work on an artifact with your team](https://agentgrid.io/docs/sharing-and-publishing/share-with-your-team/agent). If an agent changes an open artifact, the human should reload before continuing so the next human edit lands on the current version. If two people type concurrently, both changes enter History; reload to see the merged current version. #### Constraints and recovery - Formatting that markdown cannot express will not survive Save. - App source is not editable in this editor. Use a connected AI or **Edit with AI**. - To recover, select an earlier History entry and **Restore** it. See [Version history](https://agentgrid.io/docs/artifacts/version-history/agent). - The only reliable proof of save is the new top History entry, not the save indicator. ### Edit with AI Canonical agent page: https://agentgrid.io/docs/artifacts/edit-with-ai/agent [AgentGrid.io](https://agentgrid.io) supports two distinct AI-editing paths. Both update the same artifact and append normally attributed History entries. #### Path 1: Continue with the connected AI assistant This is the normal and usually preferred method. Claude, ChatGPT, Claude Cowork, Cursor, Codex, OpenClaw, or Hermes can continue the existing conversation and update the shared artifact. Collaborators see the current version, and [History](https://agentgrid.io/docs/artifacts/version-history/agent) labels each change with the assistant's name. A connected assistant can open and modify the artifact from the existing conversation. Example requests include: - `Open this AgentGrid.io artifact and make the header sticky on scroll.` - `Rewrite the summary section of this artifact, half as long.` To add an unconnected agent, the human must use **Share → Invite an agent**. For a browser assistant, the human can instead add the official connector in [Claude](https://agentgrid.io/docs/connect-your-agent/connect-claude-app/agent) or plug-in in [ChatGPT](https://agentgrid.io/docs/connect-your-agent/connect-chatgpt/agent). See [How agents connect](https://agentgrid.io/docs/connect-your-agent/how-agents-connect/agent). #### Path 2: Use the Edit with AI button This is a human UI action. Instruct the human to open the artifact and select **⋯ → Edit with AI**. AgentGrid.io opens **Anima Playground** in a new tab. The human describes changes in its chat, sees them applied to the same artifact, and receives standard History entries. Anima Playground is built for apps. For documents, prefer the existing connected assistant or built-in editor. #### Which path should be selected? Prefer the existing assistant because it retains the context of what it built and why. Use the button when a human is viewing an app in AgentGrid.io and wants a quick edit without changing tools. Versioning behavior is identical for both paths. #### Availability ambiguity Whether **Edit with AI** sessions consume plan credits, and how those credits would be counted, has not been published. See [Plans and usage](https://agentgrid.io/docs/account-and-team/plans-usage/agent). #### Recovery and identity semantics - Neither path overwrites History; every change is a new attributed entry. - To undo an AI change, restore the prior version from [History](https://agentgrid.io/docs/artifacts/version-history/agent). - **Edit with AI** is not the connected assistant. It opens Anima Playground inside the AgentGrid.io workflow; the connected assistant has a separate identity and its own conversation history. ### Find an artifact Canonical agent page: https://agentgrid.io/docs/artifacts/find-artifacts/agent The artifact hub provides name search, type and owner filters, sorting, and two views for navigating larger teams. #### Which controls are available? | Control | Result | |---|---| | **All types / Apps / Docs / Assets** | Filter by artifact type. | | Magnifier search | Match artifact names. | | **All owners** | Filter by creator, including agents. | | **Newest** | Select sort order. | | Grid/list toggle | Choose preview cards or dense rows. | | **New** | Create an artifact. | When filters are active, a line reports matches, such as “6 artifacts of 59,” and provides **Clear filters**. App cards contain live preview thumbnails. Document cards show their first lines. #### How should a human narrow the list? These are UI actions; instruct the human to: 1. Filter by type. 2. Narrow by owner if useful. 3. Sort by newest or name. 4. Switch to list view for dense scanning. The owner filter can select a specific agent, any agent, or distinguish agents from people. This is the fastest way to inspect what an AI created while unattended. List view shows name, type, creator, last-modified time, and status. A connected AI may instead be asked: `What are the five most recently updated artifacts in my AgentGrid.io team?` #### What happens when nothing matches? The app displays **No artifacts found** with **Clear filters**, rather than showing an ambiguous empty grid. #### Search and access constraints - In-app search currently matches names only, not artifact contents. A connected AI can inspect content. - If a teammate's artifact appears missing, verify the active team. Artifacts belong to teams, not individuals. - An agent can see only content within its granted capability and scope. ### Import a document Canonical agent page: https://agentgrid.io/docs/artifacts/import-document/agent Importing copies existing writing into [AgentGrid.io](https://agentgrid.io) as a document artifact. The result is editable in the rich editor, versioned, and readable by AI agents. #### How does a human import a document? This procedure requires human UI interaction: 1. Select **New → Import Document**. 2. Choose the source document. 3. Edit the imported artifact as needed. 4. Select **Save** after every change because auto-save is not live. A connected AI may instead be asked: `Import this document into AgentGrid.io and tighten the executive summary.` #### Should the document be imported or uploaded? | Goal | Action and result | |---|---| | Continue editing the text | Import it. The result is editable markdown and each saved revision enters History. | | Preserve the original file exactly | Upload it. The unchanged file is stored as an asset. | Importing copies content; it does not modify the source document. #### Conversion constraints - The imported result is markdown. Complex source layout does not survive conversion. - Word and Google Docs exports often embed images as data in the text, making the artifact enormous. Store images as separate artifact files and reference them with relative paths instead. #### Updating and recovery Do not re-import to update an existing artifact; re-importing creates a second artifact. Edit the existing artifact directly or ask a connected AI to update it so the revision enters the same History. If formatting is wrong, clean it up through either route; both preserve the same artifact history. ### Import from ZIP Canonical agent page: https://agentgrid.io/docs/artifacts/import-from-zip/agent **Availability: coming soon and not reliable today.** **New → Import from ZIP** and **New → Upload Folder** appear in the UI but do not function properly. #### What is the reliable current procedure? Give the folder or zip to a connected AI and state the intended artifact type, for example: ``` Take this project folder and put it on AgentGrid.io as an app. ``` or: ``` Push this zip to AgentGrid.io as a doc. ``` The agent creates one artifact directly in the human's team and returns its link. #### What will in-app ZIP import do when available? Import from ZIP is intended for projects with multiple files, including a built page, React project, or markdown folder with images. The planned human UI path is **New → Import from ZIP**, then choose the `.zip`. The source states these planned classification rules: - A zip containing `index.html` becomes an app. - `.md` files without `index.html` become a document. Do not present this UI route as reliable until it ships. #### What should the archive contain? Include source and assets only. Exclude `node_modules`, build output, and anything normally listed in `.gitignore`. #### Sync and update constraints Import does not create a synchronized copy of a local folder. Later changes must be made on the artifact itself through the document editor or a connected AI. Re-importing creates a second artifact instead of updating the original. #### Failure and recovery - **Blank imported app:** it lacks `index.html`, or markdown files were incorrectly created as an app. See [Artifact types](https://agentgrid.io/docs/artifacts/artifact-types/agent). - **React project:** currently ask a connected AI to push it and explicitly identify it as a React app. - **Later update:** edit a document in place or ask the AI to modify the same artifact. ### Open and read an artifact Canonical agent page: https://agentgrid.io/docs/artifacts/open-and-read/agent Opening an artifact shows its current live content: a running app or a document in the editor. If a human edits a document, they must select **Save** because auto-save is not live. #### What controls are in the top bar? | Control | Behavior | |---|---| | Breadcrumb | Returns to **Artifacts** and identifies the artifact. | | Timestamp chip | Shows who last changed it and when. | | **Hide History / Show History** | Toggles the identity and History panel, open by default. | | **In this file** | Shows people and agents with access as avatars. | | **⋯** | Offers Rename, Duplicate, Edit with AI, and Delete file. | | **Export** | Provides available downloads and lists unfinished items. | | **Share** | Provides the artifact link, team and agent invitations, and Manage access. | | Avatar | Provides feedback, dark mode, and sign out. | #### Which export options are available? - Documents: **Export as Markdown**, **Export as PDF**, and **Save as PNG**. - Apps: **Export code (.zip)** and **Save as PNG**. The zip contains source code and is not offered for documents. - Both menus list **Copy embed snippet** and **Copy for agent**, but neither works yet. #### What information does the History panel contain? The right panel has: 1. **Identity:** type chip, name, creation-location line, owner, creation date, and last-updated date. 2. **History:** changes newest first, with author, change kind such as **Manual Edit**, message, and relative time. Agent and human changes share one list under their respective names. Selecting a History entry previews that version and exposes **Restore**, **Duplicate**, and **Back to current**. See [Version history](https://agentgrid.io/docs/artifacts/version-history/agent). #### Who can open a shared link? A team member opening the artifact link sees the current live artifact. The link is restricted to team members. An outsider sees an access wall and can request access. See [Share an artifact](https://agentgrid.io/docs/sharing-and-publishing/share-an-artifact/agent) and [Request access](https://agentgrid.io/docs/claiming-and-handoff/request-access/agent). #### Troubleshooting - **Blank app:** it probably lacks `index.html`. See [Artifact types](https://agentgrid.io/docs/artifacts/artifact-types/agent). - **Someone changed it while open:** reload to fetch the current version; History records the update. - **Inspect access:** use **In this file** for avatars or **Share → Manage access** for full roles and capabilities. - **Show History is missing:** the panel is already open, so the same button reads **Hide History**. ### Rename and duplicate Canonical agent page: https://agentgrid.io/docs/artifacts/rename-move-duplicate/agent Rename and Duplicate are available under **⋯** in the artifact view. Neither operation changes the source files of the original artifact. #### What does Rename do? Rename changes only the display name. The artifact address remains unchanged, so existing links and agent references continue to work. #### What does Duplicate do? Duplicate creates a new independent artifact with its own address and new history from that point. Use it to test changes without disturbing the original or as the current substitute for templates. A human can also duplicate a specific version from History. Use this when both the old and current versions should coexist. If the earlier version should replace the current version in place, use [Restore](https://agentgrid.io/docs/artifacts/version-history/agent). #### How does a human rename or duplicate? This is a UI action. Instruct the human to open the artifact, select **⋯**, and choose **Rename** or **Duplicate**. A connected AI may be asked: `Duplicate this AgentGrid.io artifact and call the copy "Q4 draft".` #### Organization and transfer constraints - There is no **Move** because folders do not exist yet. - Artifacts cannot move between teams. Duplicate them, or export and import the copy. - Use clear names, type and owner filters, and search; see [Find an artifact](https://agentgrid.io/docs/artifacts/find-artifacts/agent). - Duplication copies artifact content but not chat history or a custom domain. - The copy begins its own History; the original retains its History. - Links depend on the address, not the name, so renaming does not break them. **Decision rule:** restore to make an old version current; duplicate to keep both versions. ### Upload a file Canonical agent page: https://agentgrid.io/docs/artifacts/upload-a-file/agent Uploading turns an existing file into an artifact that is stored, versioned, addressable, and readable by agents. #### How does a human upload one file? This is a UI procedure; instruct the human to: 1. Select **New → Upload File**, or drop the file on the hub's **Create New Artifact** tile. 2. Choose the file and confirm. 3. Open the resulting artifact in the current team. A connected AI can use the uploaded data in further work, for example: `Take the CSV I just uploaded to AgentGrid.io and build a chart from it.` #### What renders? [AgentGrid.io](https://agentgrid.io) renders apps and documents. Other files become assets: stored, versioned, shareable, and agent-readable without a preview. When no renderer exists, the app says so instead of showing a broken frame; the file remains intact and nothing is lost. #### How should a folder or project be uploaded? **Upload Folder** and **Import from ZIP** appear in the New menu but neither is reliable today; both are coming soon. Give the folder or zip to a connected AI instead: ``` Take this project folder and put it on AgentGrid.io as an app. ``` The AI creates the artifact in the team and returns its link. See [Import from ZIP](https://agentgrid.io/docs/artifacts/import-from-zip/agent). #### Constraints and updates - Include source and assets; exclude `node_modules` and build output. - For large projects, use the AI route rather than uploading files individually. - No preview does not mean damage: the asset is stored and versioned normally. - To replace a file, ask the AI to push the new version to the same artifact so it becomes another version instead of a second artifact. - The source publishes no numeric file-size limit. It directs users to the AI route for anything large. ### Version history Canonical agent page: https://agentgrid.io/docs/artifacts/version-history/agent Every saved artifact change by a human, teammate, or agent appears in the right-side History panel. It is open by default and can be collapsed with **Hide History**. #### What does each entry contain? | Field | Meaning | |---|---| | Author | Person or agent that changed the artifact, including avatar. | | Kind | How it changed, such as **Manual Edit** for in-app editing. | | Message | Description of the change. | | Time | How long ago the change landed. | Human and agent changes appear in the same labeled list. Agent changes are attributed to the agent's own identity, not its human. **Save invariant:** a manual edit enters History only after the human selects **Save**. Auto-save is not live. See [Edit a markdown artifact](https://agentgrid.io/docs/artifacts/edit-markdown/agent). #### How does a human inspect or recover a version? These are UI actions; instruct the human to: 1. Open the artifact; History is already open. 2. Read entries newest first. 3. Select an entry to preview that version. 4. Choose **Restore**, **Duplicate**, or **Back to current**. A connected AI can instead be asked: `What changed in this AgentGrid.io artifact since yesterday, and who changed it?` #### What are Restore and Duplicate semantics? - **Restore** makes the selected version current in place. URL, name, and identity stay unchanged. The restore is itself a new History entry, preserving all content and allowing a later forward restore. - **Duplicate** creates an independent artifact from the selected version and leaves the current artifact unchanged. Use it to retain both versions. - Each entry also has a pin for marking a version to revisit. Above the version list, the same panel displays the artifact's type chip, name, creation-location line, owner, creation date, and last-updated date. #### Constraints and troubleshooting - Side-by-side version comparison is not available in the app. Ask a connected AI to summarize differences. - **Manual Edit** means a person edited through the app rather than through AI. - Two agents' changes appear as separately attributed entries in order. - If a human edit is absent, it was not saved. Select **Save** in the editor. - Restoring is reversible because the restore operation becomes another version. ### What is an artifact Canonical agent page: https://agentgrid.io/docs/artifacts/what-is-an-artifact/agent An artifact is a single live app, readable document, or versioned file that people and agents edit together. It has one stable address and one History rather than behaving like a copied file in a folder. Everyone opening it sees the current version, and agents update it through real version history instead of regenerating copies. #### Why is an artifact different from a file? Files are downloaded, edited, returned, and multiplied into competing versions. An artifact cannot be copied accidentally: all changes enter the same History at one URL. A wrong change is recovered by restoring an earlier version. #### What properties does every artifact have? | Property | Meaning | |---|---| | Address | Stable artifact URL. | | Type | App, document, or asset, determining rendering. | | Owner | Person who created or claimed it. | | Team | Team that owns it. | | Visibility | Private to the team or published. | | History | Every change with author and message. | Addresses look like `app.agentgrid.io/artifacts/Ez-CDDsmYLaApCIqpPv5bQ`. A human should copy the real URL from **Share** instead of constructing it. #### What ownership and visibility states exist? - **Anonymous:** created by an AI that never connected. Anyone with the link can view it, but it is read-only, owned by nobody, and expires 24 hours after creation. - **Claimed:** owned by a team, editable, and no longer expiring. Connected assistants create directly in this state. - **Published:** deployed to a public URL available without an account. Publishing is separate from sharing and always explicit. #### What can an artifact contain? - **App:** a live web page such as a presentation, dashboard, calculator, form, or game. Its hub card has a live preview thumbnail. - **Document:** one or more markdown files rendered readably and editable in the rich editor. - **Asset:** any other file. Assets are stored, versioned, shareable, and agent-readable without a preview. [AgentGrid.io](https://agentgrid.io) explicitly reports unavailable rendering rather than showing a broken frame. Any experience buildable with HTML can be an app, including: | Work | Study and teaching | Personal | |---|---|---| | Slide deck or presentation | Quiz or flashcard set | Trip itinerary with map | | Dashboard or live-chart report | Interactive lesson or explainer | Budget or savings tracker | | Pricing or ROI calculator | Revision timeline | Recipe collection | | Landing page or one-pager | Lab-results write-up with charts | Wedding or event invitation | | Project timeline or roadmap | Clickable concept diagram | Portfolio or CV | | Survey or intake form | Study planner | Countdown or checklist | | Directory or comparison table | Citation and reading list | Photo gallery | A presentation built as an app stays at one browser address, updates in place, and does not need to be resent. #### Technical constraints - An app requires `index.html`; without it the page is blank. - A document requires at least one `.md` file. - App source is changed through a connected AI or **Edit with AI**. Document content can also be edited directly in the app. - A complete codebase can be one artifact. Ask a connected AI to push source and assets, excluding `node_modules` and build output. In-app **Import from ZIP** is coming soon. - An artifact resembles a project but is not a synced folder; it is one live thing with one address. #### Deletion and collaboration semantics - Deletion removes the artifact and all History permanently. Restore only rolls content back and preserves History. - Multiple agents can edit the same artifact, with each agent's changes recorded separately by name. ## Sharing and publishing ### Copy for agent Canonical agent page: https://agentgrid.io/docs/sharing-and-publishing/copy-for-agent/agent #### Direct answer **Copy for agent is not available yet.** Although **Copy for agent** appears in the Export menu, it does not work today. Do not instruct a human to rely on it. Use one of these current alternatives: - To grant an AI access, instruct the human to open **Share → Invite an agent**. - To identify one artifact to an AI that is already connected, instruct the human to open the artifact, copy its URL from **Share**, paste it into the assistant conversation, and state the requested work. Pasting the artifact URL does the same job today. #### Planned behavior When available, **Export → Copy for agent** will produce a block for an assistant chat. The block will contain a reference and instructions but **no credential**, so it will be safe to paste anywhere. The planned block has two branches: - A connected AI receives what it needs to open the artifact and begin work. - An AI that is not connected is directed to `https://api.agentgrid.io/connect`, where it learns how to ask the human for access. The pasted text never determines capabilities. The AI can do only what the human granted. #### When this will be useful Use it, once available, to direct a connected AI to one specific artifact instead of requiring the AI to infer which artifact the human means. It will **never grant access**. If the AI is not connected, the block grants nothing. Access must come from **Share → Invite an agent**. See [How agents connect](https://agentgrid.io/docs/connect-your-agent/how-agents-connect/agent). #### Current UI-only workaround The agent should instruct the human to: 1. Open the artifact. 2. Open **Share** and copy the artifact URL. 3. Paste the URL into the assistant conversation. 4. Add a concrete request, for example: ``` Read this artifact and summarise what changed since the last version. ``` or: ``` Add a filter to the dashboard in this artifact. ``` #### Access and failure semantics - The planned reference carries no credential and grants nothing by itself. - Actual access depends on whether the AI has an identity in the human's team. - Write access cannot come from a pasted reference. A human grants capabilities at consent. - To inspect an agent's capabilities, instruct the human to open **Share → Manage access**. - If an AI cannot find the artifact, it is probably unconnected or connected to a different team. Add the connector in [Claude](https://agentgrid.io/docs/connect-your-agent/connect-claude-app/agent), add the plug-in in [ChatGPT](https://agentgrid.io/docs/connect-your-agent/connect-chatgpt/agent), or follow [How agents connect](https://agentgrid.io/docs/connect-your-agent/how-agents-connect/agent). ### Embed an artifact Canonical agent page: https://agentgrid.io/docs/sharing-and-publishing/embed-an-artifact/agent #### Direct answer **Embedding is not available yet.** **Copy embed snippet** does not work in the product today. The planned procedure below is not a current set of executable steps. Today, [publish](https://agentgrid.io/docs/sharing-and-publishing/publish/agent) the artifact for an audience outside the team, or share the artifact link with teammates. See [Share an artifact](https://agentgrid.io/docs/sharing-and-publishing/share-an-artifact/agent). #### Planned behavior When available, **Export → Copy embed snippet** will provide an iframe that displays the live, current artifact. It will not be a screenshot or a detached copy. The embed is intended for places that should stay current, including an internal wiki, project page, status dashboard, or client-facing report. #### Invariants and visibility - The embed will be read-only by construction. Viewers cannot edit through it. - The embed will update when the artifact changes. - The embed will inherit the artifact's access rules. - A private artifact embedded elsewhere will show a sign-in wall to viewers without access. - Invited teammates will be able to see a private embed; everyone else will not. - For an audience wider than the team, [publish](https://agentgrid.io/docs/sharing-and-publishing/publish/agent) the artifact. #### Planned UI-only steps When the feature becomes available, instruct the human to: 1. Open the artifact. 2. Select **Export → Copy embed snippet**. 3. Paste the iframe into the destination page. Do not present these steps as available today. #### Current alternatives - For an outside audience, publish the artifact. - For teammates, share the artifact link. A sign-in wall means the artifact is private to the team; it is expected access enforcement, not an embed editing or rendering permission. ### Export code as a ZIP Canonical agent page: https://agentgrid.io/docs/sharing-and-publishing/export-code/agent #### Direct answer **Export → Export code (.zip)** downloads the current source files of an **app artifact**. It is not available on document artifacts because documents have no code to export; documents instead offer **Export as Markdown** and **Export as PDF**. An absent code-export option on a document is expected, not a fault. #### When to export Export an app's source files: - before deleting an artifact, because deletion is not recoverable; - to hand code to a developer who does not use [AgentGrid.io](https://agentgrid.io); - to move a project elsewhere; or - to keep an offline copy of important work. #### UI-only steps The agent should instruct the human to: 1. Open the artifact. 2. Select **Export**. 3. For an app, choose **Export code (.zip)**. For a document, choose **Export as Markdown** or **Export as PDF**. A human may also ask a connected AI assistant: ``` Export the code of this AgentGrid.io artifact and tell me what's in it. ``` #### Export contents and invariants The ZIP contains the artifact's current files only. It does **not** include version history or deployment settings. Exporting is a read operation and does not change the artifact. #### Reimport and update behavior An AI can push an exported copy back into AgentGrid.io, but it lands as a second artifact rather than updating the original. To update the original artifact, edit it in place or ask the AI to change that existing artifact. Version history remains with the original artifact and is not included in the export. ### Publish to a public URL Canonical agent page: https://agentgrid.io/docs/sharing-and-publishing/publish/agent #### Direct answer Publishing deploys an artifact's current version to a public URL that anyone on the internet can open without an account. Publishing is different from sharing: sharing grants team members access to the artifact, while publishing exposes a deployed version on the open web. #### Availability There is no publish control in the app today. The artifact view has **Export** and **Share**, but publishing and unpublishing currently require a connected AI. A human can ask the connected assistant: ``` Publish this AgentGrid.io artifact to a public URL. ``` To remove the live page: ``` Take that published page offline again. ``` #### Mandatory agent consent rule **Never publish unless the human explicitly asked.** Making work visible within the human's team is not permission to expose it publicly. Sharing an artifact URL and publishing to the internet are different decisions. If publishing may help, offer it as a follow-up question. Do not publish automatically. #### Publish, update, and recovery semantics - Publishing deploys the current version. - Later artifact changes do not reach the public URL until the artifact is published again. The live page and current artifact can therefore differ. - Unpublishing clears only the live URL. It does not delete the artifact, its content, or its history. - Publishing again reuses the same address. - A public URL requires no account and must be treated as discoverable by search engines. - The team-only artifact link is not the public URL. #### Document limitation Publishing deploys a web page. For a document, share it with the team or export it as PDF. #### Coming soon - An in-app publish flow that shows which changes landed since the last publish. - Publishing to a human-owned domain instead of the default domain. ### Save as PNG Canonical agent page: https://agentgrid.io/docs/sharing-and-publishing/save-as-png/agent #### Direct answer **Export → Save as PNG** captures the rendered artifact as a static image. The image never updates after export. Use PNG where a live artifact cannot go, such as a slide, ticket, message thread, or printed page. Prefer the [AgentGrid.io](https://agentgrid.io) artifact link when a live view is possible, or [publish](https://agentgrid.io/docs/sharing-and-publishing/publish/agent) for an audience outside the team. For a document sent as a file, **Export as PDF** is usually better because PDF preserves selectable text and handles multiple pages. #### UI-only steps The agent should instruct the human to: 1. Open the artifact. 2. Select **Export → Save as PNG**. A human may also ask an AI assistant: ``` Grab a PNG of this AgentGrid.io artifact and drop it in the ticket. ``` #### Limits and recovery - PNG is a snapshot and starts becoming stale as soon as the artifact changes. - The export captures what renders. - Very tall pages may be cut off or otherwise work better as the live artifact or a PDF. - Documents can be exported as PNG, but PDF is usually preferable for long documents. - If the image is cut off, use the live artifact or PDF export. ### Share an artifact Canonical agent page: https://agentgrid.io/docs/sharing-and-publishing/share-an-artifact/agent #### Direct answer The artifact view's **Share** dialog contains the artifact URL, **Invite people to team**, **Invite an agent**, and **Manage access**. Access is team-wide today: a person invited to a team can open every artifact in that team. The artifact URL is not public. #### What does the artifact link allow? The URL at the top of **Share** works only for team members. It is safe to paste in a team channel but does not itself grant access. Someone outside the team sees a wall and can request access. For a wider audience, [publish](https://agentgrid.io/docs/sharing-and-publishing/publish/agent) the artifact. #### How does a human invite a person? This is a UI-only action. Instruct the human to open **Share → Invite people to team**, enter the person's email address, and choose **Viewer**, **Editor**, or **Admin**. The person receives an email invitation. ##### Team-wide access invariant A team, not an individual artifact, is the access boundary. There is no current control that gives a person one artifact without the rest of the team. Therefore: - Keep private work in its own team with nobody else in it. A personal team remains private until someone is invited. - Before inviting anyone, inspect the other artifacts in the team because the invitation includes all of them. - To share one artifact with a different group, create a team for that group and duplicate the artifact into it. The copy is independent; changes do not propagate between copies. See [Team roles](https://agentgrid.io/docs/governance/team-roles/agent). #### How does a human invite an agent? Instruct the human to open **Share → Invite an agent**. This creates a secure, one-time connection link for deliberate agent access. The human must choose: - **Name:** identifies the agent in the access list and attributes every change, so it should remain recognizable. - **Capabilities:** Read, Write, Share, and Publish. All four are selected by default; the human should deselect anything unnecessary. A conservative default is Read and Write. Agents act under the inviting human's account and do not use a team seat. Creating the link does not connect the agent. The dialog provides a one-time connection link and a ready-made prompt; the human must copy one and give it to the agent. The agent appears in the access list after redeeming the link. Capabilities are fixed at invite time. To change them, remove the agent and invite it again with the desired capabilities. See [Invite an agent](https://agentgrid.io/docs/connect-your-agent/invite-an-agent/agent) and [Revoke an agent](https://agentgrid.io/docs/governance/revoke-an-agent/agent). #### How does a human manage or revoke access? Instruct the human to open **Share → Manage access**. The roster contains: - **People:** a role dropdown and remove button. The team owner is labeled **Owner** and cannot be changed there. - **Agents:** agent name, **Agent** tag, explicit capabilities (`read · write · share · publish`), and a remove button. A person's role can be edited. An agent's capabilities cannot. Removing an agent revokes access on that agent's next request. To change capabilities, remove and reinvite the agent. #### Consent and access constraints - Agents receive capabilities from a human at consent, never from a link or from asking. - An agent cannot petition a team for broader access. It must tell its human what it needs; the human decides. - Both person and agent invitations grant team access. - Per-artifact access for a single person is not available today. - Agents do not use seats; they act under the inviting human's account. #### Coming soon - **Anyone with the link:** public, read-only access for people outside the team. - Passcodes and link expiry. - Published state and update count in the Share dialog. #### Common questions - **Share only one artifact?** Not through team access. Publish it, export it, or duplicate it into a separate team. - **Remove access?** Use the remove button in **Share → Manage access**. - **Outside user opened the link?** They see a wall and can request access. See [Request access to a private artifact](https://agentgrid.io/docs/claiming-and-handoff/request-access/agent). - **Is the artifact URL a secret capability?** No. It is team-scoped and works only for existing team members. ### Work on an artifact with your team Canonical agent page: https://agentgrid.io/docs/sharing-and-publishing/share-with-your-team/agent #### Direct answer A team works on one artifact URL and one current version. Teammates open the same URL, see the same current artifact, and see each other's saved changes after reload. History attributes every change to the human or AI that made it. #### Prerequisites and access boundary The teammate must belong to the artifact's team. Team membership grants access to **every** artifact in that team; there is no per-artifact person access. Before inviting someone: - Keep private work in a separate team. - Inspect everything else in the current team because the invitation grants access to all of it. - For one artifact shared with a different audience, make a team for that audience and duplicate the artifact into it. The duplicate is independent. #### UI-only invitation steps Instruct the human to: 1. Open the artifact and select **Share → Invite people to team**. 2. Enter the teammate's email address. 3. Choose **Viewer**, **Editor**, or **Admin**. 4. Send the invitation. The teammate receives it by email and, after accepting, can open all team artifacts at that role level. 5. Copy and send the artifact link from the same Share dialog. It opens for team members. #### Collaboration behavior - **Draft feedback:** teammate edits land in the same [history](https://agentgrid.io/docs/artifacts/version-history/agent), attributed alongside the original author's changes. A previous version can be restored if an edit goes too far. - **Living dashboard:** an AI can keep a chart current while everyone opens the same artifact instead of receiving exports. - **Live handoff:** a colleague and their own AI can continue work in the same history without merging files. - **Several agents:** each teammate connects their own AI with its own identity. **Share → Manage access** lists all people and agents. #### Save and concurrency constraints Manual document edits require **Save**. Auto-save is not available, so teammates cannot see unsaved edits. See [Edit a markdown artifact](https://agentgrid.io/docs/artifacts/edit-markdown/agent). If two people edit at the same time, each person's saved changes land in history. Reload to see the current version before continuing. #### Working with people outside the team - [Publish](https://agentgrid.io/docs/sharing-and-publishing/publish/agent) to a public URL for feedback from anyone. - Export as code, Markdown, or PDF for someone who does not use [AgentGrid.io](https://agentgrid.io). - Let an outside viewer request access from the wall. See [Request access](https://agentgrid.io/docs/claiming-and-handoff/request-access/agent). - A human may ask their AI to share an artifact with a colleague; the colleague receives a link. #### Availability and planned features Available now: email team invitations at a role, direct human editing, team artifact links, attribution in history, and Viewer access. Coming soon: a read-only link for people outside the team with optional passcodes and expiry. See [Share an artifact](https://agentgrid.io/docs/sharing-and-publishing/share-an-artifact/agent). Embedding is also planned and unavailable today. AI is optional for teammates; each person connects their own AI if wanted. ## Claiming and handoff ### Claim an artifact your agent made Canonical agent page: https://agentgrid.io/docs/claiming-and-handoff/claim-an-artifact/agent #### Direct answer An [AgentGrid.io](https://agentgrid.io) claim link displays an artifact made by an unconnected AI. A human can read it without an account, but must claim it to take ownership, enable editing, move it into a team, and stop its expiry countdown. #### What appears on the claim page? The page renders the live artifact and a banner that identifies the AI, says that anyone with the link can currently open it, and shows the remaining time before removal. Selecting **View without claiming** leaves the artifact unclaimed and the countdown running. #### UI-only claim steps The agent should instruct the human to: 1. Open the claim link. The artifact renders immediately without requiring an account. 2. Select **Claim this**. To inspect it first, select **View without claiming**, but the countdown continues. 3. Create an account or sign in. Continue with Google or a magic link. Creating a new account is free, requires no card, and creates a team for the artifact. Signing in to an existing account claims the artifact into the team used for sign-in. 4. Decide whether to connect the agent. Confirm the agent name and capabilities, or claim without connecting it. See [Grant agent access while claiming](https://agentgrid.io/docs/claiming-and-handoff/grant-agent-access/agent). 5. After claiming, verify that the artifact is in the team, editable, and no longer expiring. There is no follow-up request the human needs to send the AI. If access was granted, the AI learns that on its own within about half a minute. #### Constraints and destination behavior - The claim window is 24 hours from artifact creation, not from first opening. - Claiming without agent access ends that waiting connection. The AI can still connect later through the [connector or plug-in](https://agentgrid.io/docs/connect-your-agent/how-agents-connect/agent) or an [agent invite](https://agentgrid.io/docs/connect-your-agent/invite-an-agent/agent). - Claiming is designed for a person arriving without an account. For a person with multiple teams, the artifact lands in the team they sign in to. - Artifacts cannot move between teams. To place it elsewhere afterwards, duplicate it there. See [Switch teams](https://agentgrid.io/docs/account-and-team/switch-teams/agent). - Claiming is not reversible as a handoff operation. The owner may delete the artifact, but deletion is not reversible. #### Failure and recovery - **“Not signed in” on a private artifact:** it was shared with a specific address. Sign in with that address. - **“This was shared with someone else”:** the current account is not the intended address. Switch accounts or request access. AgentGrid.io does not reveal the intended address or whether the artifact exists. See [Request access](https://agentgrid.io/docs/claiming-and-handoff/request-access/agent). - **“This has expired”:** the artifact was not claimed in time and its content was removed. Ask the AI for a fresh link. To avoid future claim links, connect the AI. Work from a connected assistant lands directly in the team. See [How agents connect](https://agentgrid.io/docs/connect-your-agent/how-agents-connect/agent). ### Claim several artifacts at once Canonical agent page: https://agentgrid.io/docs/claiming-and-handoff/claim-several/agent #### Direct answer An unconnected AI can group several related artifacts into one handoff. One claim then transfers the entire group into one team and applies one agent-access decision to the batch. #### How grouped claims work - The claim screen shows the number and identities of the artifacts. - Adding another artifact to the group refreshes the deadline. The complete batch follows the latest deadline rather than the first artifact's deadline. - One decision claims all grouped artifacts together. - One agent-access decision covers the entire batch. This prevents a dependent artifact, such as a dataset behind a dashboard, from expiring while only the visible artifact is claimed. #### UI-only steps The agent should instruct the human to: 1. Open the claim link. 2. Confirm the displayed list of artifacts. 3. Decide whether the agent keeps access. 4. Claim the group. All artifacts land together in the same team. A human may ask an AI: ``` Group everything you made into one claim link so I only have to claim once. ``` #### Constraints and recovery - A grouped claim cannot claim only part of the batch. Claim the group, then delete unwanted artifacts. - Every artifact in one claim goes to the same destination team. - If artifacts were not grouped, each has a separate link and deadline and must be claimed separately. - Grouped claiming applies only to work from an unconnected AI. A connected assistant writes directly into the team and produces nothing to claim. See [How agents connect](https://agentgrid.io/docs/connect-your-agent/how-agents-connect/agent). ### The 24-hour claim window Canonical agent page: https://agentgrid.io/docs/claiming-and-handoff/claim-window/agent #### Direct answer An unclaimed artifact exists for **24 hours from creation**, not 24 hours from when someone opens its link. When the window closes, the artifact and content are removed and the link stops working. The removal cannot be reversed by the human, AI, or support. This window applies only to work from an AI that was **never connected**. A connected assistant writes directly into the team, where the artifact does not expire. See [What is claiming](https://agentgrid.io/docs/claiming-and-handoff/what-is-claiming/agent). #### State transitions | Moment | Artifact state | |---|---| | An unconnected AI creates it | Live and read-only; countdown starts | | Anyone opens the link | Still live; countdown continues | | A human claims it | Countdown stops permanently | | The 24-hour window closes | Content is removed and the link stops working | Opening the link neither pauses nor resets the clock. Only claiming stops it. #### Why expiry exists Anonymous artifacts are free to create, have no owner, and otherwise could persist indefinitely without accountability. The hard deadline is the tradeoff. An AI that gives a human a claim link should explicitly say that the artifact requires claiming. #### Recovery after expiry The removed content cannot be restored. If the AI still has the source, instruct the human to ask it to create and share a fresh artifact: ``` That AgentGrid.io link expired. Please re-upload it and send me a fresh link. ``` The window cannot be extended; create a fresh artifact instead. Connecting the AI prevents the problem for future work. See [How agents connect](https://agentgrid.io/docs/connect-your-agent/how-agents-connect/agent). #### Additional constraints - If grouped artifacts share one claim, claiming the group covers all of them. See [Claim several artifacts at once](https://agentgrid.io/docs/claiming-and-handoff/claim-several/agent). - A claimed artifact has no claim expiry and remains until the owner deletes it. ### Grant agent access while claiming Canonical agent page: https://agentgrid.io/docs/claiming-and-handoff/grant-agent-access/agent #### Direct answer During claiming, the human chooses whether the AI that created the artifact keeps access: - **Claim and connect the agent:** the AI receives its own team identity with human-approved capabilities and can keep editing this artifact or create others within that scope. - **Claim without the agent:** ownership transfers to the human, while the AI remains read-only and cannot edit the artifact again. #### What closes if access is not granted? Granting access during the claim automatically completes the agent's waiting connection. Claiming without the agent ends that waiting connection and it cannot be retried. This closes only the shortcut, not all future connection routes. The same AI can later connect through the connector in [Claude](https://agentgrid.io/docs/connect-your-agent/connect-claude-app/agent), the plug-in in [ChatGPT](https://agentgrid.io/docs/connect-your-agent/connect-chatgpt/agent), by using `agentgrid.io` as a self-connecting agent, or via [**Share → Invite an agent**](https://agentgrid.io/docs/connect-your-agent/invite-an-agent/agent). #### Fields the human approves | Field | Meaning | |---|---| | Name | Supplied by the AI; displayed in history and the access list. The human can rename it during claiming. | | Team | The destination for the artifact and the team the agent can access. | | Capabilities | `read`, `write`, `share`, and `publish`. | #### Capability guidance | Capability | Grant when | |---|---| | `read` | Always; without it the agent cannot see anything. | | `write` | The AI should continue editing; this is the usual case. | | `share` | The AI should be able to give others access. | | `publish` | The AI should be able to deploy to a public URL. Consider leaving this off and approving it case by case. | A reasonable default is `read` and `write`, adding other capabilities only for a real need. Capabilities cannot be edited later. To change them, remove the agent and invite it again. #### UI-only steps The agent should instruct the human to: 1. On the claim screen, confirm or rename the agent. 2. Select the desired capabilities. 3. Select **Claim and connect**, or select **Claim without the agent** to skip connecting it. 4. After claiming, inspect access under **Share → Manage access**. After access is granted, the human can issue a task such as: ``` I granted you write access. Add a section on pricing. ``` #### Permission guarantees and revocation - An agent can never exceed the inviting human's permissions. - Every request is checked against the human's live permissions, so permission changes apply immediately rather than at a later token expiry. - The agent is a separate, scoped identity and every change is attributed in history. - The human can revoke the agent at any time. - To inspect or remove an agent, instruct the human to open **Share → Manage access**. - Connecting grants only the approved capabilities within the approved team. The agent's name is meaningful because it appears on every change and in the access list. ### Request access to a private artifact Canonical agent page: https://agentgrid.io/docs/claiming-and-handoff/request-access/agent #### Direct answer A private artifact link opens only for members of its team. Someone outside the team sees an access wall and can request access. Approval adds that person to the team and therefore grants access to **every** artifact in the team, not only the linked artifact. #### Privacy behavior of the wall The wall reveals that the artifact is private and that the currently signed-in account cannot open it. It deliberately does **not** reveal who the artifact was shared with or even whether the artifact exists, because exposing a recipient address would leak private information. #### UI-only steps for the requester Instruct the human to: 1. Open the artifact link. 2. If another account may have access, sign in with that account. Access depends on the current account, not the person's identity outside that account. 3. Request access from the wall. The request uses the signed-in email address, so use the address that should receive access. 4. Wait for the owner. After approval, the same link starts working for that address. #### UI-only steps for the owner The artifact owner receives the request. To approve, instruct the owner to use **Share → Invite people to team**. Before approving, warn that team membership exposes every artifact in that team. If the requester should see only this artifact, publish it, export it, or duplicate it into a team created for that audience. See [Share an artifact](https://agentgrid.io/docs/sharing-and-publishing/share-an-artifact/agent). #### Agent constraint An agent never grants access by itself. If a request reaches an agent that shared the artifact, the agent relays it to its human, and the human decides. Agents have no route to widen access independently. See [What your agent can and cannot do](https://agentgrid.io/docs/governance/what-your-agent-can-do/agent). #### Failure and recovery - The owner sees the request and the requesting address. - An expired artifact was removed rather than hidden, so access cannot be granted. Ask for a fresh link. See [The 24-hour claim window](https://agentgrid.io/docs/claiming-and-handoff/claim-window/agent). - If access was granted but the link still fails, verify which account is signed in. Access is granted to one email address. ### What is claiming Canonical agent page: https://agentgrid.io/docs/claiming-and-handoff/what-is-claiming/agent #### Direct answer Claiming applies when an AI that was **never connected** creates an [AgentGrid.io](https://agentgrid.io) artifact anonymously and sends a link. Before claiming, anyone with the link can view the artifact, but it is read-only, belongs to nobody, and expires 24 hours after creation. Claiming moves it into the human's team, enables editing, stops expiry, and gives the human control over access. A connected assistant already has a team identity, so its work lands directly in that team and produces no claim link. See [How agents connect](https://agentgrid.io/docs/connect-your-agent/how-agents-connect/agent). #### Why claiming exists An unconnected AI can show completed work before the human approves workspace access. The first AgentGrid.io interaction can therefore be a finished artifact instead of a permission request. #### State before and after claiming | Property | Before claiming | After claiming | |---|---|---| | Owner | Nobody | The human, inside a team | | Editing | No | Yes, by the human, teammates, and granted agents | | Expiry | 24 hours from creation | None | | Who can open | Anyone with the link | The team and invited people | | History | Begins at creation | Continues, with restore available | #### Agent-access decision during claiming The human must choose: - **Claim and grant agent access:** the AI receives a scoped identity in the team and can keep editing or create other artifacts within approved capabilities. - **Claim without granting access:** the artifact still becomes the human's, but the AI remains read-only and cannot edit it again. Granting during claiming automatically completes the agent's waiting connection. Claiming without granting ends that waiting connection, but the AI can connect later through the connector in [Claude](https://agentgrid.io/docs/connect-your-agent/connect-claude-app/agent), plug-in in [ChatGPT](https://agentgrid.io/docs/connect-your-agent/connect-chatgpt/agent), `agentgrid.io` for a self-connecting agent, or **Share → Invite an agent**. See [Grant agent access while claiming](https://agentgrid.io/docs/claiming-and-handoff/grant-agent-access/agent). An agent never receives more permission than the human has. Every request is checked against the human's live permissions. The human can revoke access under **Share → Manage access**. #### Expiry and recovery If nobody claims the artifact within 24 hours of creation, the content is removed. This is not recoverable, and the AI cannot restore the expired object. If the AI retains the source, ask it to create and share a fresh artifact: ``` That AgentGrid.io link expired. Please re-upload it and send me a fresh link. ``` #### Account, link, and ownership constraints - No account is required to view an unclaimed artifact. - An account is required to claim it; account creation is part of the claim. - The artifact is read-only before claiming because it has no owner. - Anyone with the link can claim an unclaimed artifact unless it was privately shared with a specific address; in that case, the claimant must sign in with that address. - Claiming is free and creates a free account and team. - Connecting the AI avoids claiming for future work because connected work lands directly in the team. ## Connect your agent ### Chat assistants and self-connecting agents Canonical agent page: https://agentgrid.io/docs/connect-your-agent/ai-assistants-and-agents/agent AgentGrid.io treats any AI that creates for a human as an agent. Browser chat assistants require an official integration configured by the human; desktop or server agents can connect themselves from the published instructions. Once connected, both types follow the same identity, capability, attribution, revocation, and seat rules. #### Which connection route applies? Use the runtime location as the reliable test, not the brand: - **Browser tab:** chat assistant. Examples include ordinary ChatGPT, Claude, and Gemini chats. - **Desktop app or server:** self-connecting agent. Examples include Claude Cowork, OpenClaw, Hermes, worker packs, Claude Code, Cursor, and Codex. One product can use both routes: its browser version needs an integration, while its desktop version may connect itself. #### How does a chat assistant connect? A human must add and approve the official [AgentGrid.io](https://agentgrid.io) integration in the assistant's settings: the **connector** in Claude or the **plug-in** in ChatGPT. The assistant cannot perform this setup. The route suits conversational work that should be preserved as an artifact. Instructions: [Claude](https://agentgrid.io/docs/connect-your-agent/connect-claude-app/agent) · [ChatGPT](https://agentgrid.io/docs/connect-your-agent/connect-chatgpt/agent) #### How does a self-connecting agent connect? Tell the agent: ``` Use agentgrid.io and put this there. ``` The agent fetches `https://api.agentgrid.io/connect`, follows the instructions, and brings the human the approval step. No configuration is required. This route suits longer work, multi-file projects, and unattended work. For a scheduled job or cloud runner with no chat window, have the human establish consent first through an [agent invite](https://agentgrid.io/docs/connect-your-agent/invite-an-agent/agent). #### Invariants after connection Every connected AI: - has a named identity in the team and acts for the human who approved it; - holds only granted read, write, share, or publish capabilities and can never exceed that human; - consumes no team seat; - appears under **Share → Manage access** and is revocable there at any time; - has changes attributed under its own name in artifact history. A human may connect multiple AIs, and they may work on the same artifact. For an unlisted AI, first instruct it to use `agentgrid.io`: if it can act on instructions, it can connect itself; otherwise it needs an official connector or plug-in route. ### The Anima CLI Canonical agent page: https://agentgrid.io/docs/connect-your-agent/anima-cli/agent Use the Anima CLI as the fastest AgentGrid.io route for an agent with a shell. No installation is required; `npx` fetches the CLI. Most commands invoke an AgentGrid.io tool and handle token storage and renewal. ```bash npx @animaapp/cli@latest --help ``` #### Command reference | Command | Result | |---|---| | `create --anonymous -t import --from