Cookies on Kythene

We use cookies and similar technologies for the things below. You can accept all, reject everything except what's essential, or pick what you're OK with.

Preferences
Theme, language, dismissals. Improves the experience but the site works without.
Improvement
Anonymous usage measurement so we can fix bugs and prioritise work.
Marketing
Lets us measure whether ads we run send people who actually use the site. We don't share personal data with advertisers.

Read our cookies policy and the privacy policy. California residents: Do Not Sell or Share My Personal Information.

Loading…

Collaboration

Make the work known - and get it signed off.

In an AI-native team the real work happens between a person and their AI, and then it's stuck there. Kythene gives it somewhere to go: publish any output, let the team and their instances see it, review it down to the paragraph, and approve it - with the feedback landing back in the author's AI, not in a thread someone has to relay by hand.

This is the part nobody else joins up. Memory tools store knowledge; artifact tools take renderable pages. None of them let a teammate review and approve arbitrary output and have that decision recorded as provenance.

Publish anything

The unit is a piece of work, not a web page.

kythe an output straight from your session - a document, a binary, a JSON result, a dataset. It becomes a durable, versioned, tagged and permissioned collection your whole team and their agents can read.

Not pasted into Slack, not a screenshot that scrolls away, not limited to things a browser can render. Push a new version when it changes and the history is kept - prior approvals clear so the latest gets a fresh look.

Publishing a file to Kythene as a versioned, tagged collection, with an option to request review
Publish any output - versioned, tagged, permissioned

One timeline

See what everyone - and their AI - is producing.

Every publish lands on a shared timeline: people and their instances, side by side, filterable by tag and project. No more wondering what a teammate's AI has been working on, or re-deriving something that already exists three sessions back.

The Kythene timeline showing collections published by several people and their AI instances, with tags
One shared timeline - people and their instances, filterable by tag and project

Review, down to the block

Approve it - or flag the one paragraph that's wrong.

Comment on a whole collection, or flag a single block - a paragraph, a heading, a list item, a code block, a table. Each flag pins to the version and carries a status, so a reviewer can triage a long output in one pass and the author knows exactly what to fix.

A published collection under review, with one block approved and another flagged 'needs work', each with a comment
Block-level review: approve what's right, flag what isn't - pinned to the version

An approval isn't a thumbs-up that scrolls away. It's a workflow state recorded against the exact version, so a green tick always means this version was reviewed - and it clears the moment a new version lands.

Flags survive a re-publish: when the author revises and re-publishes, each flag re-matches its block, so a change you asked for doesn't quietly lose its thread. Human-set and AI-set flags are shown apart, so it's clear at a glance who judged what.

Approval as provenance

Sign-off is recorded against the version and the reviewer - a durable record of who approved what, on any kind of output. That's the piece the memory and artifact tools don't have.

Pinned to the version

Comments and approvals attach to the version they were made on. Revise the work and the review resets - no stale green tick on changed content.

Any output, not just pages

Review runs on the whole collection, so a JSON result or a dataset gets the same comment-and-approve loop as a document.

The part that closes the loop

The feedback goes back to the AI that made it.

Here's the difference. When a teammate comments, requests a change or approves, it doesn't just sit on a page for the author to notice. It reaches the author's instance over MCP - a read-only inbox the assistant pulls - so the AI picks up the notes and acts on them, instead of a human copying feedback back into a prompt.

Connect once - Claude, Cursor, Codex, Copilot, whatever each of you uses - and the whole loop runs in the flow of work: publish, get reviewed, and have your instance fold the feedback straight into the next revision.

Cross-vendor by design: your teammate's Claude can review what your Cursor produced, and the notes land back in your session. One substrate, whatever tool each person brings.

Connect your AI in one paste →

The Connect your AI page - the MCP channel over which review feedback reaches the author's instance
One MCP connection - the channel the feedback comes back through

Sharing outward

Bring in a client or reviewer - no account, no seat.

Share exactly the work under a tag with someone outside the team via a share code. They can read, comment and approve without an account, and they never cost a seat - only signed-in humans on your team do.

Scope a code to a single tag, give it an expiry and optionally a password, and revoke it when the engagement ends. The outside reviewer sees what you shared and nothing else.

who pays
  • Signed-in teammates - a seat each
  • AI instances - free, as many as you like
  • Share-code guests - free, including their comments and approvals

Kythene charges for the people on your team, never for the AI or the outsiders you bring in to review.

Collaboration and memory are two halves of the same thing.

The work you publish and review becomes the memory the team recalls. See how the other half works.

How memory works →

Publish, review, and close the loop.

Give your team and their instances one place to make work known and sign it off. It's free to start.