Real-time collaboration — your team and your AI, on the same page.

Live cursors. Threaded comments. CRDT sync that holds together offline. And the agent you mentioned answers in the thread — not in a chat panel, not a side summary — the way a human peer would, through the same protocol Claude reads your notes. AI suggestions land block by block with diff-before-apply, like AI Co-Edit in the editor; every keystroke survives in version history.

Collaborators per doc
7
Capability areas
8
Merge conflicts by design
0
Real-time
The cast

Seven seats at the same table.

A Knovya note is a room. You sit at it — and so do six AI agents you already work with. Each one keeps its own identity, its own voice, its own tone preferences.

  • You Author
  • Claude Reasoning
  • Cursor Coding
  • ChatGPT Drafting
  • Gemini Long context
  • Copilot Pair-coding
  • Goose Local agent

Word Copilot can edit. Notion AI can suggest. Knovya treats the agent as a co-author — with a name, a thread, and a place in the document's history.

Experiment 02 · The Lab

Watch the room. Press replay.

Forty-five seconds of one note, three collaborators. You start a paragraph. Cursor finishes a code block. You ask Claude for a stronger thesis — and Claude replies in the thread.

Q3 plans auth migration — v3

Auth migration — v3

Why does this matter? — thesis paragraph, in progress…

scene 1 · you are typing

Comments · mentions · AI replies in thread — included on every plan, including Free. Live cursor presence · workspace activity feed — part of Team workspaces.

Eight capabilities · what holds the room together

Multiplayer is more than cursors.

A complete real-time stack covers eight surfaces — from the keystroke layer to the conflict resolver. Each one is here.

Block-anchored comments

Threads attach to a block, not a character offset — so they survive edits, paragraph splits, and reordering. Note-level threads cover the whole document.

Every plan

Mention everything

Mention a workspace member, an AI agent, or another note. The agent or person responds in-thread; the linked note becomes a live reference.

Every plan

Live presence & awareness

Named cursors with throttled updates, a clean disconnect when a tab closes, and an avatar cluster that shows who else is in the room right now.

Team workspaces

Multi-agent provenance

Hover a block to see who edited it last. The note-level breakdown shows the share by author across the past 7, 30, or 90 days — humans and AI on equal footing.

Every plan

In-thread AI replies

Mention @claude in a comment and the response appears as another comment, not a separate chat. Three layers of loop protection keep threads from running away.

Every plan, scaled by tier

Suggesting mode

Block-level propose · accept · reject for humans and AI alike. A side-by-side diff before anything lands. Race-safe acceptance under concurrent reviewers.

Pro & Team

Slash AI in-line

Press / in any block and four AI commands surface: rewrite, summarize, translate, explain. They land as suggestions or apply directly — your call.

Pro & Team

Smart Merge resolution

When concurrent edits genuinely collide, a four-pane modal opens: auto-merged blocks above, conflicting blocks paired side by side, with one-click take all yours or theirs.

Every plan

Real-time was built for humans.
The room got bigger.

Google Docs solved live editing in 2006 — for two humans typing at once. Notion added it. Figma made it visual. The model has not changed in twenty years.

But the room you write in today has more than humans. Claude drafts the first paragraph. Cursor closes the code block. ChatGPT proofs the conclusion. The collaboration model has not caught up to who actually shares the doc.

The gap
Word Copilot edits in private. Notion AI answers in a side panel. Google Docs has no place for an AI agent at all.
The fix
Stop treating AI as an assistant. Give it a seat, a name, and a thread.
The lineage

Sixty years from one demo to seven collaborators.

Live editing did not arrive in 2026. Five turning points led here.

  1. 1968
    Engelbart — The Mother of All Demos A live, multi-cursor editing session over a network — almost everything we still call collaborative software was demonstrated in ninety minutes. SRI · first multi-user editor
  2. 1989
    Ellis & Gibbs — Operational Transformation A central server orders every edit and rewrites operations to keep documents consistent. The algorithm Google Docs would still rely on twenty years later. ACM SIGMOD · OT paper
  3. 2011
    Shapiro et al. — CRDTs Conflict-free Replicated Data Types — data structures that mathematically converge without a server arbitrating order. Real-time editing without a single point of truth. INRIA · CRDT formalization
  4. 2019
    Figma — OT to CRDT switch Six months of engineering to swap collaboration engines: thirty percent less server load, offline-first multiplayer, and the proof that CRDT scales to production. Figma engineering blog
  5. 2026
    Knovya — multi-agent multiplayer CRDT for keystrokes. Threads for conversation. Provenance for authorship. Six AI agents as named co-authors. The first room where humans and AI write together as equals. ★ Knovya · production
First of its kind

Nobody else lets the AI live in the doc.

Every notes app added "collaboration" in the same shape: humans editing, AI on the side. Knovya is the first to seat the agent at the table — in the comment thread, with a name, with provenance, with the same conflict-free sync as a human peer.

  • Google Docs humans · OT · no AI in thread
  • Notion humans · AI in side panel
  • Word + Copilot humans · AI edits silently
  • Obsidian humans · via plugins only
  • Figma humans · canvas multiplayer
  • ★ Knovya humans + 6 named AI agents · in thread
Surfaces

Where the room shows itself.

Four moments where multi-agent multiplayer becomes visible — in the editor, in the thread, in the history, in the collision.

Live editor multi-cursor

Three cursors in the same paragraph. Each carries a name label, a color, and a throttled update stream. Disconnect cleanly — the cursor disappears.

Comment threads in-thread AI

Mention @claude and the agent answers as another comment — same thread, same scroll position. Persona-aware to your tone.

Provenance breakdown who wrote what

Hover the title bar — a horizontal bar shows the share. AI agents earn a row. The same view across 7-, 30-, or 90-day windows.

Smart Merge when edits collide

CRDT auto-merges most things. When two edits genuinely conflict, this view appears — pick the side that survives, block by block.

Bonded with

Real-time composes with the rest of Group III.

19
Block Editor
22 block types. Every block is a real-time unit of edit.
21
Templates
Open a template; collaborate on it. Same room, structured start.
25
Version History
Every keystroke, restorable. Provenance keeps the author trail.
22
Share
Public read, viewer-only links, expiry dates. The room, opened.
Frequently asked

A few honest answers.

What is real-time collaboration in Knovya?
Humans and AI agents — Claude, Cursor, ChatGPT, Gemini, Copilot, Goose — can edit the same note simultaneously. Live cursors show who is where, comment threads accept @-mentions of any participant, and the AI you mention replies in the thread itself. The whole stack is CRDT-backed, so concurrent edits never collide.
How is this different from Google Docs?
Google Docs uses Operational Transformation, which requires a central server to order every keystroke and breaks down on long offline sessions. Knovya uses a CRDT — edits merge mathematically without coordination, so offline work syncs cleanly when you reconnect. The bigger difference: Google Docs is for humans only. Knovya treats AI agents as peers in the document.
Can AI agents really reply inside a comment thread?
Yes. When you write @claude in a comment, Claude responds in the same thread — not in a separate chat panel, not as a side panel summary. The reply appears as another comment, persona-aware to your tone and language preferences, with three layers of loop protection so threads cannot run away. Mentions work for every connected agent (Claude, Cursor, ChatGPT, Gemini, Copilot, Goose) through MCP.
What happens when two people edit the same paragraph?
Most of the time, nothing — the CRDT merges concurrent edits into a single consistent state. When the edits genuinely conflict (the same span of text deleted and rewritten in two different ways), Knovya opens a Smart Merge view: auto-merged blocks on top, conflicting blocks paired side by side for you to pick from.
Does collaboration work offline?
Yes. Edits made offline merge cleanly when the device reconnects. The CRDT under the hood treats your edits as a partial truth that can rejoin the larger truth without a central server arbitrating order. This is the same algorithm Figma adopted in 2019 to ship offline-first canvas multiplayer.
Can I see who wrote what — including which AI agent?
Every block carries a provenance signal. Hover any paragraph to see who edited it last. The note-level breakdown shows the share over the past week or month — for example, you 60%, Claude 32%, Cursor 8%. AI agents are first-class authors, not anonymous helpers — see version history for the full author trail.
Which plans include real-time collaboration?
Comment threads, mentions, AI agent replies, Smart Merge resolution and version history are part of every plan, including Free. Live cursor presence and the workspace activity feed are part of Team workspaces, where multiple members share one space. Pro adds suggesting mode and slash AI in-line for higher-throughput AI use.

Open the room. Bring the cast.

Comments, mentions, and AI replies are free, forever. Live presence comes with a Team workspace.

Start free See Team
element 20 · Group III — Editor