Glacier — User Guide
Glacier is a flow-based workspace: a Kanban board and doc editor in one place, built for small teams where AI is already part of the daily workflow.
This guide covers the web app. If you want to connect Claude as an AI collaborator, see the MCP Guide.
Getting started
1. Sign in
Go to getglacier.ai and sign in with your email or Google account. If you've been invited to a workspace, accepting the invitation automatically places you in the right one.
2. Your workspace
A workspace contains all your projects, boards, and docs. Everything is scoped to it — you only see what belongs to your team.
3. Projects
A project has a Kanban board and a doc section. Select a project from the sidebar to open it. Each project has its own columns, cards, and linked documents.
The Kanban board
Glacier's board is flow-based, not sprint-based. Work moves left to right through columns. The goal is continuous flow, not timed sprints.
Columns
Each column has a type:
| Type | Meaning |
|---|---|
| Queue | Work waiting to be started |
| Active | Work in progress |
| Waiting | Blocked or waiting on something external |
| Done | Completed |
WIP limits
Active columns can have a WIP (Work in Progress) limit. If a column is over its limit, it's signalled visually. WIP limits are a core lean practice — they surface bottlenecks before they become crises.
Cards
A card is a unit of work. Each card has:
- Title — what the work is
- Description — freeform markdown text
- Priority — low / medium / high
- Assignee — who owns it
- Subtasks — a checklist of smaller steps
- Labels — colour-coded tags
- Linked docs — reference documents attached to the card
- GitHub links — linked issues or PRs with live status
- Attachments — images and PDFs uploaded directly to the card
Creating a card
Click + at the top of any column, or use the keyboard shortcut if available. Give it a title and press Enter. Open the card to fill in description, priority, and other fields.
Moving a card
Drag and drop a card to a different column to move it through the flow. Cards can also be moved from the card detail panel.
Attachments
Cards support file attachments: images (PNG, JPG, WebP, GIF up to 10 MB) and PDFs (up to 25 MB).
To attach a file:
- Click Add in the Attachments section of a card, or
- Drag and drop a file directly onto the card detail panel
Images appear as thumbnails. Click a thumbnail to open the full image in a lightbox. PDFs appear as a file row with a download link.
To remove an attachment, hover over it and click the × button. This unlinks the file from the card (it does not permanently delete it).
Via MCP: Claude can upload attachments using the upload_attachment tool, then link them with add_attachment_to_card. Attachments are included in the output of get_card.
Card IDs
Every card gets a short ID like GLACIE-5 or SKILLS-12. These are stable and useful for referencing cards in conversations with Claude or in commit messages.
Docs
Docs live inside projects and can be linked to cards. They're useful for specs, runbooks, meeting notes, or anything that needs to stay close to the work.
Creating a doc
Open the Docs section of a project and click New doc. The editor supports standard markdown formatting: headings, bold, italic, lists, code blocks, and horizontal rules.
Inline images
Drag an image file onto the editor, or paste one from the clipboard, to upload it and insert it inline. Images are stored in the same secure storage as card attachments and are only accessible to workspace members.
Linking a doc to a card
From the card detail panel, you can attach an existing doc. The doc then appears in the card's linked documents section and is readable by Claude via MCP.
Workspace-level docs
Docs can also exist at the workspace level — not tied to any project. These are good for team-wide references like onboarding guides or working agreements.
Annotations & suggestions
When an AI agent reviews a document, it can leave suggestions and comments directly inside the text. You review them without leaving the doc.
Annotated text
Text with an annotation attached appears marked in the document. The visual style tells you what kind of annotation it is:
- Strikethrough text + inline replacement — a suggestion. The original text appears crossed out and the proposed replacement follows it in the same line.
- Underlined text — a comment. No replacement, just a note anchored to that text.
Solid marks are from teammates; dashed marks are from AI agents.
The Annotations panel
Click the annotation icon in the top-right toolbar to open the side panel. The panel lists every open annotation grouped by author — agent suggestions first, then your own, then other reviewers.
Each entry in the panel shows:
- The original text and the proposed replacement, for suggestions
- The rationale for the change
- How long ago it was created
- Accept and reject buttons
The hover popover
Hover over any annotated text to see a compact popover with the suggested change and rationale. You can accept or reject directly from the popover without opening the panel.
Accepting and rejecting suggestions
Accept — applies the suggested replacement. The original text is replaced and the annotation disappears.
Reject — dismisses the suggestion without changing the document.
For comments, the equivalent action is Resolve — it closes the comment thread without any text change.
Bulk accept / reject
When two or more suggestions are open, Accept all and Reject all buttons appear at the top of the Annotations panel.
- Accept all — applies every open suggestion in document order.
- Reject all — dismisses all open annotations (suggestions and comments). An undo notice appears for 8 seconds — click it to restore them all if you change your mind.
Agent edit policy
By default, AI agents connected via MCP edit documents directly. You can change this at the workspace, project, or document level.
Go to Workspace Settings → Agents to set the workspace default. Individual projects and documents can override it.
| Policy | What it means |
|---|---|
| Strict | Agents can only propose suggestions. Direct edits are never allowed. |
| Allow override | Agents default to suggestions but can request direct edits. |
| Direct (default) | Agents edit directly by default. Suggestions can still be requested. |
The setting on a document takes precedence over the project, which takes precedence over the workspace.
Flight levels (optional)
Flight levels are an optional way to organise a board into multiple layers — for example, a strategic level and an operational level. They're opt-in and disabled by default. New projects always start as flat Kanban boards.
If your project uses flight levels, use the level selector in the board header to switch between views.
Account settings
Access your account settings from the bottom of the sidebar. From there you can:
- Update your profile name and avatar
- Generate or rotate your MCP API key (for Claude Desktop / Claude Code)
- View your current workspace
Tips
Keep cards small. A card should represent work completable in a few days at most. If it's bigger, break it into sub-items.
Use the description. A card with a clear description is something Claude can read and act on via MCP. Sparse cards create sparse context.
Respect WIP limits. If a column is full, finish something before starting something new. That's the point.
Link docs to cards. When a card has a spec or reference document, link it. It keeps context together and makes it available to AI agents.