AI-assisted SQL, signed and bundled.

Double-click the file or drag it into a browser window. The canvas boots showing the title, drop zone, and just the Tree tab.

Select all the XMLs in features/ and drag them onto the canvas drop zone at once. New tabs appear: Streams, Identities, Sign, Transforms, Endpoints, Compose, Schemas, Console, Bundle.

Tree · Streams · Identities · Sign · Transforms · Endpoints · Compose · Schemas · Console · Bundle

Optional but recommended — gives name attribution on the tour author's chunk signatures rather than just a fingerprint.

stylesheets/xsl-tone-terse.xml
stylesheets/xsl-tone-public.xml
stylesheets/xsl-tone-annotated.xml

The Transforms tab now shows three transforms; chunk regions will have three tone-readers available.

Stream declaration first, then the chunks. Order of chunk drops doesn't matter — the streams feature uses each chunk's seq attribute to order them.

Five green ticks. If any reds, refresh the canvas and re-drop.

Oracle Live SQL is a free hosted Oracle 23c playground. You'll use it as your execution engine in chunk 4.

Why a separate tab? Oracle Live SQL blocks iframe embedding (X-Frame-Options: SAMEORIGIN), and even if it didn't, embedding a full DB UI inside the canvas would be the wrong architectural separation. Canvas organizes; Oracle executes.

Click "Get Started" if you don't have an account. Oracle asks for email + a chosen password; no payment info needed for the free tier.

Keep this tab open — you'll come back to it in phase 5.

To flick tones on any chunk: click the chunk in Tree, then click a different tone reader radio in the Inspector. Same content, three voices.

source id: sql-q1.xml
number:    SQ1
title:     SQL question
register:  lay
paragraph: (your question)

SEND TO TREE. The new source appears in Tree.

Click into sql-q1.xml in the Tree. Drill until you find the paragraph element. Its Inspector shows an address like mr://sql-q1.xml/0.0.0 — copy that exact string.

Stub from chunk 3:

<endpoint id="sql-author" provider="anthropic" model="claude-sonnet-4-20250514" max-tokens="800" prompt-ref="mr://sql-q1.xml/0.0.0">
  <system>You are an Oracle SQL expert. Given a plain-English question, write the Oracle SQL that answers it. Use Oracle Database 23c syntax. Include short comments. If multiple queries are needed, separate them clearly. Do not invent schema — ask for clarification if too vague.</system>
</endpoint>

Update prompt-ref to your actual question address before processing.

The badge shows "invoking" for a few seconds, then a new <response> element appears as a child of the endpoint with the proposed SQL inside.

If the SQL isn't quite right, edit your question region in the Tree and click INVOKE again — the response will reflect your refined question.

Click on the response to render it. Select-all + copy the SQL portion. If the response has prose around the SQL, copy just the code block.

Oracle's result panel appears below the editor with the returned rows. If you see errors, that's also useful — you can capture them and ask the AI to revise in a follow-up endpoint.

source id: sql-r1.xml
number:    SR1
title:     SQL result
register:  lay
paragraph: (the Oracle output — rows and/or error messages)

SEND TO TREE. You now have your three artifact sources: question (sql-q1), proposed SQL (in the endpoint source), real result (sql-r1).

An ECDSA-P256 keypair generates locally. Private key in localStorage; public key embedded in your future signatures.

For each: pick the source from the dropdown, leave path blank, click SIGN. A <signature> element appends as a child of the source root.

Optional: sign the endpoint source too. That gives a verifier proof you authorized the AI call.

The bundle is one file containing your whole session: features, sources, signatures, content hashes. A recipient with a canvas drops it and recovers everything.

Reuse for any external service: Jupyter, GitHub, Figma, your IDE. The canvas captures the artifact of each external operation as a signed addressable region. The bundle ships the whole session.