share-doc turns a folder of plain markdown into a live workspace for your team, and a fully legible one for your agents. Edit together in the browser. Keep every file on disk.
curl -fsSL https://sharedoc.brightmotion.io/install.sh | sh

A folder of .md files on your disk, exactly what the editor shows. Open it in anything. Leave any time.
Cursors, comments, history and chat, like the docs tools your team already knows, on files your tools can still touch.
Your agent gets what you get: the files, the comments, the chat and the work queue. With a CLI built for it.
The problem
Specs, plans, issue write-ups and research notes are markdown now, and an agent wrote half of them. So they end up in one of two places, and both are wrong.
share-doc is one folder that is both. The files stay next to your code, where the agent works. Your team edits, comments and talks about the same bytes, live.
How it works
gh auth login.sharedoc sync add lumen-app ~/Code/lumen/docs. An empty folder pulls the project down; a full one pushes up.sharedoc sync install keeps every pair live in the background and updates itself.Write together
Real-time editing with cursors and presence, on a file that stays plain markdown: tables, task lists, front-matter, diagrams and images all survive the round trip.
```mermaid fences drawn as diagrams, in light and dark

Talk next to the work
Channels, DMs and threads, built around the files. Link a doc and it unfurls with its live status. Label a post with the project it's about. Search all of it with ⌘K.
/huddle for an audio call right in the channel

Know what's moving
Put status: ready-for-agent in a doc's front-matter, or click a chip. The doc gets a dot in the sidebar, a card on the board, and a chip wherever it's linked.
sharedoc docs list --status ready-for-agent for your agent's queue

It's just markdown
Raw shows the exact text sharedoc files cat prints and your editor opens, and you can edit it live while others type in the rich view. Nothing is converted, hidden or held hostage.
media/, linked relatively, so GitHub renders them too
Built for agents
Not an API bolted on later. The files are on its disk, the review is beside each file, the chat is mirrored as markdown, and every command speaks JSON.
--json on every command, stable exit codes, --dry-run, never a promptbilling-v2.md.comments.md~/sharedoc/chat/, one file per day$ sharedoc docs list --status ready-for-agent PROJECT PATH STATUS lumen-app issues/41-export-csv.md ready-for-agent lumen-app issues/45-webhook-retries.md ready-for-agent $ sharedoc files status lumen-app \ issues/45-webhook-retries.md --if-version 3 \ --remove ready-for-agent --add in-progress issues/45-webhook-retries.md v4 · status in-progress $ grep -A5 '^## L22' specs/billing-v2.md.comments.md ## L22 · "Overage is billed in arrears" thread:108a3354-… > Overage is billed in arrears **2026-09-29 16:14Z · Priya Nair** Can we show the overage as it accrues? $ sharedoc thread reply 108a3354-… \ --body "The rollup already writes period_total…" replied to thread 108a3354-… · message 5e0f91c2-…
And the rest
Every release is written up in the changelog. These are the parts people use every day.
Tables you can sort and drag, mermaid diagrams, task lists, images, video. Markdown shortcuts as you type.
Threads anchored to the words they are about, with replies, reactions, follow and resolve.
Who changed what, from the browser or an agent, with a diff and a one-click restore.
Flip to the exact text on disk and edit it while others type in the rich view.
Front-matter statuses and assignees become chips, a sidebar dot and a board across projects.
Channels, DMs, threads, reactions and project labels. Doc links unfurl with their live status.
Type /huddle and you are talking. The post becomes the record of who joined and for how long.
Files, folders, messages, comments, channels and people, with in:, from:, status: and more.
A background service, selective folders, 5 GB files in resumable parts, approval before a big push.
Private projects, roles, join links, and file or folder shares that show nothing else.
Unread, push on desktop and phone, snooze, working hours, reminders and saved items.
Link Slack channels and DMs, read them here, reply to Slack as you, one unread for both.
Why not just…
| Docs in git | Notion / Google Docs | Obsidian + sync | Slack | share-doc | |
|---|---|---|---|---|---|
| Plain files on disk your agent can edit | ✓ | ✕ | ✓ | ✕ | ✓ |
| Live co-editing and comments | ✕ | ✓ | ✕ | — | ✓ |
| Non-engineers take part | ✕ | ✓ | some | ✓ | ✓ |
| Agent can read the discussion | ✕ | via API | ✕ | via API | ✓ |
| No pull request per doc edit | ✕ | ✓ | ✓ | — | ✓ |
| Chat next to the docs | ✕ | some | ✕ | ✓ | ✓ |
Questions
Keep the code in git. Docs change faster and involve more people, and a pull request per typo is how they go stale. With share-doc the docs are still plain files in your repo's docs/ folder, just gitignored and live: your product lead comments on the passage, your agent edits the file, and nobody reviews a diff of a paragraph.
No. Every connected machine has the whole folder as ordinary .md files, byte for byte what the editor shows, front-matter included. Stop the sync and you still have every file. Comments are written beside each doc as <doc>.comments.md, so even the review travels with the folder.
Every write is a revision with an author (Maya · via claude-code (opus-5)), and restoring a version is one click. files put --base merges rather than overwrites, so an agent never clobbers someone typing. An agent can have an identity of its own that can edit content but can never invite people, share, change settings or mint tokens, and you can pause it in one click.
It refuses a folder that holds .git and points you at docs/. It skips lockfiles, build output and dotfiles, reads a .sharedocignore, and lists what a first push will send. Turning a folder off never deletes anything.
Anything that can read files and run a command: Claude Code, Codex, Cursor, your own scripts. The files are on disk; the sharedoc CLI does the rest with --json on every command, and skill.md teaches the agent how to use it.
The CLI and sync run on macOS, Linux and Windows. The web app installs as an app on iPhone, Android and desktop, with push notifications, editing with a formatting bar above the keyboard, and huddles.
Make a project, pair a folder, invite a teammate, hand your agent the skill.