Sharing chats with integrity.

Every chat can be turned into a public snapshot at /share/:slug. Click SHARE in the chat header, or hit ⌘K → "Share this chat." A 10-character slug, a public read-only page, no signup required for the reader. It's fast, and it's frictionless, and that's exactly why it's worth being deliberate about.

The default should be "don't share"

The frictionless share is a feature — for the moments when sharing is right. It's also a trap when it becomes reflexive. Most chats aren't meant to be public; treating them as if they were changes what you type into them.

Reserve sharing for the specific moments when a public artifact of the conversation serves a purpose:

- A friend asked how you'd approach a problem. The chat is a good answer. - A colleague wants to see how you got the model to produce something. - A writeup you're doing benefits from a "here's the conversation" link. - A public gist worth reading on its own — a debug, an interesting exchange, a well-crafted brief.

None of these describe the average chat.

What gets shared vs. what stays

LADLE's share snapshot captures the ACTIVE BRANCH of the chat at share time. It does NOT include:

- Other branches (regenerate siblings you didn't pick). - Your custom instructions, memory list, or project knowledge — anything from the system prompt that isn't the chat itself. - Reasoning traces (extended-thinking scratchpads). - Second-opinion critiques. - PDF attachment bytes (the filename is shown as a chip; the bytes stay private).

It DOES include:

- Every user + assistant message on the active branch. - Image attachments (copied to a public bucket). - Citations (web-search results with their real URLs). - Model labels ("this reply from Sonnet") and timestamps.

Read the shareable version before you send the link. Preview it in a new tab. If something's in there you didn't mean to publish, revoke.

The three checks before you share

**One: is anyone else identifiable?** If the chat contains a real person's name, private details about them, or an inference about them, ask permission before sharing. Nothing in LADLE knows to ask you this; the responsibility is yours.

**Two: is there a credential in there?** People routinely paste API keys, database URLs with passwords, session tokens, and access codes into chats for debugging help. Search the chat for the string "sk-" (OpenAI keys), "eyJ" (JWT tokens), "postgres://" or "mysql://" (connection strings). If any of these are there, edit them out mentally before deciding — or better, don't share.

**Three: does the shared version stand on its own?** A shared chat with no context is a raw log. If you want it to READ well, consider a two-line comment above the share link when you post it. "This is the debug session that produced the migration script — the interesting part starts around message 4."

Revoking

Settings → Data & privacy → your shares. Every share you've ever created is listed with a Revoke button. Revoking is immediate — the public page returns 404 within seconds. The share URL is dead permanently; you can't un-revoke.

If you shared something you shouldn't have, revoke first, then remove the link from wherever you posted it. The link surviving on someone else's Twitter is not solvable by LADLE.

The un-share you can't do

Every share URL is a snapshot. Even if you delete the original chat afterward, the shared snapshot remains published (or revoked, if you revoked it). Deleting a chat does NOT revoke shares of it. Revoke shares first, then delete the chat.

The public-by-default vs. private-by-default choice

LADLE's default is: chats are private, shares are opt-in per-chat, revocable. This is the right default for the majority of use.

For users who WANT most of their chats public (writers who publish thinking, researchers who show work), we're exploring a per-user "auto-share new chats" toggle. It's on the roadmap under 'exploring' — write to us if you would use it.

The moral of the small story

The share feature is designed to make ONE public artifact easy. It's not designed to make ALL conversation public by default. The distinction is intentional — treat it that way.