new
Present shows a diagram full screen
Present, on a diagram's page in the dashboard and on every row of the Diagrams list, shows the diagram full screen, fitted and enlarged to the screen on its own canvas color, for showing on a video call: the page presents the version shown, colored when Color what changed is on, and a row presents the latest version. F toggles full screen, + and − zoom, Shift+1 fits, Shift+0 shows actual size, the mouse wheel, a pinch or a drag zoom and pan, the left and right arrow keys (or PageUp and PageDown, as a clicker sends) step through versions, C turns the change colors on and off, and Escape leaves, back to the list as it was, or to the diagram's page at the version you presented last. The controls and the cursor hide when idle, and a team's viewers can present too, with no share link needed.
new
A link opens a diagram straight into presenting
https://zindex.ai/dashboard/diagram?id=<id>&present opens a diagram straight into presenting its latest version, and adding &revision=<n> presents that version, so an agent can hand a person the link, as the Zindex skill describes. It opens for anyone in the workspace that holds the diagram, after a switch to it when another workspace is open, and a visit signed out comes back to it after sign-in. With no click behind it, the browser keeps the view inside the window, so press F for full screen.
new
Shared and published diagrams have Present and Download
A published diagram's page or a share link's page (zindex.ai/d/<ref>) has Present, so anyone with the link can show the diagram full screen and step through the versions the link shows, with the same keys and controls as on a diagram's page in the dashboard, with no sign-in needed. Its address plus ?present, such as https://zindex.ai/d/<ref>@3?present, opens it straight into presenting (press F for full screen), Older and Newer above the picture step through the versions the link shows, and Download, beside Report this diagram, saves the version shown as PNG or SVG in place of the Open full size links. Here and on a diagram's page in the dashboard, Present keeps Color what changed on at the first version, which has nothing before it to color, so stepping forward colors again and leaving Present keeps the page's box checked.
new
Shared and published diagrams show what changed
Color what changed, on a published diagram's page or a share link's page and when presenting it, colors each element added, changed or removed since the version before, as on a diagram's page in the dashboard, and a line counts them. The line gives counts only, and nothing says which fields changed or who changed them; on a share link an element can be marked changed when what changed isn't in its image. The privacy policy and terms, updated 2026-10-23, say that a share link also shows which elements each version added, changed and removed, links made before this change included, so you can revoke any that shouldn't show it.
new
Version links color what changed with no key
A version link's image for an exact version takes ?from=<revision> with no key, as in https://zindex.ai/d/<ref>@2.svg?from=1 or GET /v1/public/diagrams/{ref}@2.svg?from=1, and draws it with what changed since that version colored, and GET /v1/public/diagrams/{ref}@{revision}/changes?from={revision} returns { from, to, added, changed, removed } as counts only. from must be another version the link shows and the version an exact one (@N), or the answer is a 400, and a published diagram's links from diagram_render take it too. The colored image is drawn on each request within the 5-second budget for drawings without a key and counts toward the no-key rate limit, and both answers are cached for up to an hour, privately for a share link.
new
Each element is named in the SVG
Every SVG drawn from 7.1.0 puts each element's markup in a <g data-zindex-element="<id>"> group, so a page or a script can find an element by its id, as in svg.querySelectorAll('[data-zindex-element="api"]'), and the Rendering docs page says what each element's groups hold. Previews, ?engine=current and ?from= images, and a version's stored image whose engine is 7.1.0 or later carry the groups, while a stored image drawn earlier never changes and a PNG has none. An element's id is visible to anyone who has the SVG, share link recipients included, as the privacy policy says, so don't put in an id what you wouldn't put in a label.
new
Dark previews adapt a light design's colors
Previews in the dark, blueprint and sketch themes (the playground's theme menu, or theme on POST /v1/diagrams/preview) keep the look of a diagram designed on a light canvas. Each color the scene sets keeps its hue with its lightness flipped: white panels take the canvas color, pastels become deep tinted surfaces, containers sit below their cards, faint panels stay visible, text reaches 4.5:1 and a line keeps its authored contrast, up to 3:1. Images and AWS icons keep their colors, a scene designed on a dark canvas is drawn as written, and saved images and the images diagram_render returns are drawn with no theme, so they don't adapt.
new
Sequence diagram: after and before place messages and lifelines
Sequence diagram: after or before, the ID of the element to go next to, puts a new lifeline, actor, message or note from createNode or createEdge between two others instead of last, and moves one from updateNode or updateEdge; the element named must exist when the operation runs (ORDER_TARGET_NOT_FOUND), and other elements refuse both fields (ORDER_NOT_SUPPORTED). A message or note placed next to one inside a fragment joins it and naming the fragment places it just outside, as the Operations docs page describes. The outline lists messages and notes in the order they're drawn, so you can choose the ID, and a moved element counts as changed in the edit's changed, the history, the compare and a diff image.
new
Sequence diagram: Activation bars move with their messages
Sequence diagram: an activation bar (sequence.activation) names the messages it runs between, extensions.startsAt and extensions.endsAt, and is drawn on the lifeline startsAt arrives at, from that arrow to endsAt's, or one row tall without endsAt, so it moves with its messages as rows are added, moved or deleted. A message to or from that lifeline in those rows ends at the bar's edge, and a self-message leaves and returns on its right side, so every arrowhead shows. A bar saved with a box and no startsAt is still drawn, over the messages its box covers on the nearest lifeline, and saved versions keep their stored images.
new
Sequence diagram: Checks flag misplaced operands and activation bars
Sequence diagram: an operand whose startsAt isn't a message or note in its fragment after the first row is refused (FAMILY_SEQ_OPERAND_NOT_FOUND), and so is an activation bar whose startsAt or endsAt isn't a message in the diagram (FAMILY_SEQ_ACTIVATION_ANCHOR_NOT_FOUND). A fragment whose children skip a row between its first and its last warns FAMILY_SEQ_FRAGMENT_SPLIT and draws that row inside it, and an endsAt that comes before startsAt warns FAMILY_SEQ_ACTIVATION_ENDS_BEFORE_START and draws the bar one row tall. Each one says what to change, as the Validation rules docs page lists.
new
Signing in offers the teams you can join
After you sign up or sign in, with a password, an emailed verification code, Google or GitHub, Zindex offers each team workspace your work address can join on the way to where you were going: Join for a domain set to Auto, one click with no approval, or Ask to join where an owner approves. Not now puts an offer off in that browser, and the step is skipped on the way to an invitation, an invite link or an app's consent page, which have their own. It replaces the join section on the dashboard, and the Members page in Settings describes Auto the same way, one click offered at sign-in.
new
A workspace menu on every page
Everyone signed in has a workspace menu next to the Zindex logo on every page, to switch workspaces, join or ask to join a team your address can join, and create a team workspace, with Settings and Sign out at its foot. It replaces the switcher that sat beside the dashboard and settings tabs, and takes Settings and Sign out from the header, which keeps Dashboard.
new
Creating a team workspace is paying for it
A team workspace is paid for when you create it: name it at zindex.ai/settings/new-team, choose monthly or yearly, and pay for 2 seats, the minimum, in Stripe Checkout, at $16 a month or $160 a year. It exists once payment completes, with you as its owner and its payer, and you land on its Members page to invite people, while leaving checkout before paying creates nothing. The workspace menu, the pricing page's Team buttons and Billing all lead there, where before a team workspace could be created unpaid and sat on Free, where nobody could join it, and the menu offered a second one it then refused.
new
Connected apps move between your workspaces
A connected app such as Claude, ChatGPT or Cursor moves to another of your workspaces with one click and stays signed in: open that workspace from the workspace menu, then choose Move here in the Connected apps list on the dashboard's API keys page, which shows your apps from every workspace with the workspace each works in; an app already connected to the open workspace has no Move here. A moved app keeps its access, which your role in its new workspace narrows as it does a key's: read only while you're a viewer, writing again if you become a member. Every page that says you're in a team, whether you joined by domain, invite link, invitation or an owner's approval, offers to move your connected apps into it, while an API key stays in the workspace it was made in.
new
Connecting an app offers to join your team
When your work address lets you join a team workspace set to Auto, the page where you allow an app to connect offers to join it first, above the workspace choice. Joining there chooses the team as the workspace the app works in, so the app connects to the team in the same sign-in.
new
Approved join requests email the person who asked
When an owner approves your request to join a team workspace, Zindex emails you what your role there allows, with a link that opens the workspace and offers to move the apps you connected to your other workspaces. When the owner made you a member but the seat couldn't be added, the email says you joined as a viewer until an owner makes you a member, and a decline sends nothing.
new
Endpoints list and move your connected apps
GET /v1/oauth/grants lists your connected apps in every workspace you're in, the open workspace's first, each with its workspace (id, name and kind), where before it listed the open workspace's alone. POST /v1/oauth/grants/{id}/move with { workspaceId } moves one to another workspace you're in, from a signed-in session with a verified email, and answers { app } as listed. The app keeps its access and replaces your earlier connection of the same app in that workspace, as the Workspaces docs page describes.
new
Zindex records MCP calls, never what they send
Zindex keeps a record of each MCP call made with your sign-in or key: when, the account and workspace, the app's name, the kind of message and the tool's name, how it ended and how long it took. It never records what a call sent or the IP address. Entries are kept 90 days, and closing your account or deleting a workspace removes them; the privacy policy, updated 2026-10-22, has the details.
new
The Diagrams list shows who changed each diagram
In a team workspace, the dashboard's Diagrams list shows who saved each diagram's latest version under its Last changed time. People are named as History names them: their name, else their email, or "former member" for a closed account.
new
Diagrams saved today show a Today badge
A diagram whose latest version was saved today, in your own time zone, shows a green Today badge and the time in place of the date on the Diagrams list. What was worked on today stands out at a glance.
new
updatedBy names who saved each diagram
In a team workspace, items from GET /v1/diagrams (including ?key=) and MCP diagram_find carry updatedBy, naming who saved the latest version: their name, else their email, or former member for a closed account. It's left out in a personal workspace, and when Zindex doesn't know who saved it.
new
The homepage shows teammates' agents sharing one diagram
The homepage tells how a team's agents build on one shared diagram: an example architecture diagram saved three times by three teammates' agents, in Claude, Cursor and Copilot in VS Code. It shows each version, the diagram_create or diagram_edit call that saved it and what it changed, colored, and beside the history the stale edit Zindex refused with REVISION_CONFLICT, saving nothing, until the agent planned again against the newer version. Every picture, the example ER schema and sequence diagram under Use cases included, opens full screen in Present.
new
A Security page says how to report problems
https://zindex.ai/security says how to report a security problem, by email to support@zindex.ai with "Security" in the subject, and what happens after you report. It also lists the good-faith research Zindex welcomes and what the page covers: zindex.ai and its subdomains, including the API and the MCP server, the @zindexai/mcp package and the Zindex plugin. https://zindex.ai/.well-known/security.txt points to it, and the footer and llms.txt link it.
improved
Signed-in MCP tool calls get their own limit
Signed-in MCP tool calls draw on their own allowance per person in each workspace: 300 at once on Free, refilling 60 a minute, and 600 at once on Pro and Team, refilling 120 a minute. An agent's burst of work goes through instead of meeting the per-minute request limit. Other MCP messages, such as initialize and tools/list, have a separate ceiling of 600 a minute.
improved
Rate-limited tool calls answer as tool results
A signed-in tool call past its limit answers as a tool result with code: "RATE_LIMITED" and retryAfter instead of an HTTP 429, so the agent can wait and carry on. The call did nothing, so making it again is safe.
improved
MCP confirmation requests no longer count as failures
When an MCP client can't show a confirmation form, the first call of diagram_delete, or of diagram_share with publish, link, report or feedback, answers as an ordinary tool result without isError, carrying code: "CONFIRMATION_REQUIRED", what will happen and a confirm token as before. The client no longer counts a request to confirm as a failed call. Nothing is done until the person agrees and the call is made again with the token.
improved
Agents add .zindex.json or send feedback when wanted
An agent reads a diagram's ID and revision from .zindex.json when a repository has one and updates it after edits, and adds one only when you want the diagram tracked with the code. When the docs and the engine disagree, or something is missing, it says so, and sends Zindex feedback with diagram_share action feedback only when you want Zindex told, after you confirm it as before. The MCP server's instructions and the Zindex skill say so, and the MCP tools docs page says it of feedback.
improved
Agents include diagrams in artifacts, slides and PDFs
An agent includes one version's PNG in a page or file it publishes, such as a Claude artifact, slides or a PDF, since such pages often can't load images from other sites, and captions it with that version's address on a published diagram's page or a share link's page, such as https://zindex.ai/d/<ref>@3, or with the diagram's own address for readers in its workspace. A version never changes, so the copy never goes stale and its caption leads to every version, but the copy goes wherever the page goes, so a private diagram's, shared by link or not, belongs only in pages for people who may see it, and revoking a share link leaves the copies in place. The steps are in the REST API docs page's table of where to put a diagram, the MCP tools docs page and the Zindex skill, and diagram_render's description says such pages need the PNG file itself.
improved
Shared and published diagram images load faster
A published diagram's or a share link's image, such as https://zindex.ai/d/<ref>@3.png or GET /v1/public/diagrams/{ref}@3.svg, is kept in Zindex's edge cache close to whoever views it, so later views nearby load faster: an exact version's for up to an hour, and the latest version's, at the address without @N, for up to a minute. A share link's images stay private to every other cache, and unpublishing, revoking a share link, deleting the diagram, closing an account or deleting a workspace removes them from the cache at once, while a copy that can't be removed expires within the hour. The privacy policy, updated 2026-10-23, says what the edge cache keeps and for how long.
improved
Edge labels sit inside their frames
An edge label sits inside the frame its edge runs through, where before it was pushed into the gaps between frames, often half over a node or across a border. Labels keep clear of node outlines and frame titles where the edge allows, and when an edge has no clear place, a label can still cross a frame's border, cover a title or overlap a node, as the Layout Engine docs page describes. Every new image places labels this way in every theme, and saved versions keep their stored images.
improved
Edge labels keep clear of other edges' lines
When crowded edge labels are spread apart, a label moves only to a place clear of nodes, other labels and other edges' lines, and stays where it was if there is none. Auto-layout prefers arrangements where no label covers another edge's line and no two unrelated edges run along each other, though a label with no clear place on its edge can still cover a line. Saved versions keep their stored images.
improved
Sequence flows and UI transitions enter mid-side
A BPMN sequence flow (workflow.sequenceFlow) or a UI flow transition (uiflow.navigate, uiflow.back, uiflow.action) enters a box at the middle of the side it meets, as BPMN modelers draw them, and a flow whose source is less than 16 px off the middle runs straight in. Two such edges running both ways between the same two boxes run as two straight parallel lines through the overlap of the boxes' facing sides, unless the two boxes line up within 12 px, and message flows, edges into frames and every other edge type keep entering straight across where their boxes line up. New renders and new versions route this way, and saved versions keep their stored images.
improved
Diff colors stand out on every canvas
A visual diff, from a ?from= image or the Color what changed checkbox on a diagram's dashboard page, picks each change color by its contrast with the canvas: a deeper green, amber and red on a light canvas, where the bright ones were faint as lines on white, and the bright ones on a dark canvas. A removed element's ghost draws at 60%, no fainter than what didn't change, a removed lifeline's foot bar is dashed like its head, and the legend's Removed key is as faint as the ghosts. Removed elements are drawn where they were, so a sequence diagram's removed message keeps its row and a removed lifeline its column instead of moving to the end, and a fragment stays around its messages, ghost rows included.
improved
Verifying your email returns you where you were
Every prompt to verify your email, such as Verify now on the dashboard banner, emails you a fresh code, so a code that expired after sign-up never holds you up. Once you enter it, Zindex brings you back to the page you were on.
improved
Navigation runs Product, Teams, Examples, Docs and Pricing
The navigation at the top of every page runs Product, Teams, Examples, Docs and Pricing, where Product and Teams open those parts of the homepage, and on a phone the menu closes when you pick a link. Signed out, Sign in sits beside the button Connect your agent, which opens the Connect page where Get started opened sign-up. Connect and Contact are in the footer.
improved
The pricing FAQ says what immutable means
The FAQ on the pricing page answers "What does immutable mean?": nothing in a diagram's history can be edited, as each change adds a new version whose hash covers the one before, so any tampering would show, and you can delete a whole diagram but never rewrite its past. "Can I leave?" says nothing locks you in, on every plan: any version downloads as SVG, PNG, Mermaid or JSON, sequence diagrams as SVG, PNG or JSON, or your whole history in one file, and the scene format is an open spec. With them, the pricing page answers every question the homepage used to.
improved
The dashboard opens on your diagrams
Signing in, unless you were on your way to another page, and the Dashboard link at the top of every page land on the Diagrams list, and /dashboard redirects there. The tabs run Diagrams, API keys, Usage and Recently deleted, and the former Overview is the Usage tab at /dashboard/usage, with this month's usage against your plan's limits and Your feedback, the feedback you or your agents sent with its status.
improved
Viewers learn an owner can make them members
A viewer in a team workspace reads on the Diagrams list that they and the agents they connect can read its diagrams but not change them, and that an owner can make them a member. The playground's note for a viewer points to the workspace menu at the top of the page to switch to Personal and save there.
improved
Team workspaces without a subscription add no seats
A team workspace whose subscription ended can't have a viewer made a member or owner, or ownership handed over in a way that adds a seat, until an owner subscribes it again (409 SUBSCRIPTION_REQUIRED), as nobody new can join it. People already in it keep their roles and keep working within Free's limits, and owners can still make members viewers, remove people and hand over ownership when that adds no seat. Billing says what an unsubscribed team can't do, and on Members a viewer's Member and Owner choices are grayed out with the reason beside them.
improved
A diagram's page shows the picture first
A diagram's page in the dashboard shows the picture right after the title, with Present, Older, Newer and Color what changed above it, and the brief below it, cut to three lines with Show more. Open full size is gone from above the picture, since Present replaces it, and the image files stay under Download.
improved
A diagram's actions sit beside its name
A diagram's page in the dashboard puts Present, Share, Download and More beside the diagram's name, with a line under the name that says who can see it. Share opens one dialog for who can see it, the share links and putting it in a doc, issue or README; Download lists the version's files and all versions; More holds Report a problem, Move to a team workspace and Delete. The rest of the page is for reviewing: the picture, what changed, the brief and History, with As of beside Older and Newer.
improved
Share is a bottom sheet on phones
On phones, a diagram's Share dialog is a bottom sheet: full width at the foot of the screen, as tall as what it holds up to 90% of the screen, and a tap on the dimmed page above it closes it. Report a problem still fills the screen.
improved
Every report form fills the screen on phones
On phones, Report a problem in the playground, and Report this diagram on a published diagram's page or a share link's page, fill the screen, as Report a problem on a diagram's page in the dashboard does, so the form has the whole screen above the keyboard instead of a small box. Report this diagram keeps its title and close button at the top as you scroll. Confirmations, such as a diagram's Delete and the playground's Publish, stay centered boxes, and menus stay drop-downs.
improved
A phone's home screen shows the Zindex icon
Saving zindex.ai to an iPhone's or an Android phone's home screen shows the Zindex app icon, the logo mark in amber on the site's dark background, in place of a screenshot of the page or the small browser-tab icon. A page saved that way, such as a share link's page, still opens that page.
improved
Sequence diagram: Fragments are laid out with their messages
Sequence diagram: a combined fragment (alt, opt, loop, par and the rest) runs from its first message or note to its last, with a header row for its operator and guard, follows rows added above, inside or below it, and is as wide as what it holds plus 16 px each side, so a nested fragment sits 16 px inside its parent. Its own layout is ignored, where before a fragment with a box kept it as rows moved out from under it, and every operator draws a dashed divider with its guard above each further operand, not only alt. Fragments nest or follow each other but can't cross, a fragment outside a sequence diagram keeps its layout, and saved versions keep their stored images, as the Rendering docs page describes.
improved
Sequence diagram: Fragment guards draw over the lifelines
Sequence diagram: a fragment's operator tab, the guard after it, its dividers and each operand's guard are drawn over the lifelines, activation bars and arrows, each guard on a knockout of what is under it, so no lifeline runs through guard text. The fragment's box is still drawn under them, and saved versions keep their stored images.
improved
Sequence diagram: Fragment guards line up clear of activation bars
Sequence diagram: every guard in a fragment, the one after the operator's tab and each operand's under its divider, starts in one column after the tab, at least 38 px inside the fragment's edge, and the fragment widens when a long guard needs the room. An operand's guard no longer runs across an activation bar on the fragment's leftmost lifeline, where before it made the bar look broken, though a guard long enough to reach a bar further right still draws over it on its knockout. Saved versions keep their stored images.
improved
Sequence diagram: Edits keep operands and bars with their messages
Sequence diagram: a message or note placed with after or before joins the operand of the row it goes next to, so one placed before an operand's first message starts it. Deleting or moving a message or note that an operand or activation bar starts or ends at moves it onto the nearest row it can use, and one left with nothing is removed, with the notes FRAGMENT_OPERAND_REMOVED and ACTIVATION_REMOVED in the edit's warnings. A deleted fragment's rows stay in the fragment around it, as the Operations docs page describes.
improved
Sequence diagram: The outline shows operands and activation messages
Sequence diagram: the outline diagram_get and GET /v1/diagrams/{id} return by default gives each fragment's operator and the message or note each further operand starts at, with its guard, and each activation bar's messages, as in busy sequence.activation from m1 to m4. An agent can pick the IDs after, before and startsAt take without reading the whole scene.
fixed
Changing an edge's type changes its arrowheads
An edit that changes an edge's edgeType changes its arrowheads too, so a message made sequence.async gets an open arrowhead, where before it kept the old type's filled one. Only the arrowheads you set in endpoints that differ from the type's are kept, so view=dsp shows only those, and a version saved earlier draws as before and isn't listed as changed by the next edit.
fixed
Endpoint anchors place an edge's ends
An endpoint's anchor places that end of the edge, where before it was checked but didn't change the drawing, and its offset is optional: { "side": "left" } fixes the side and the engine chooses where on it, while { "side": "left", "offset": 0.25 } fixes the place too, 0 to 1 along the side and kept 8 px from its corners, as the Rendering docs page describes. An anchor wins over the centered dock of a BPMN sequence flow or UI flow transition and over an org chart's fixed sides, but nodes don't move for it, so an edge anchored on the side facing away from the other node routes around its node. A sequence message, a self-loop and an end that sets column ignore their anchor, with the warning EDGE_ANCHOR_IGNORED.
fixed
Edges anchored to one side stay apart
An anchored edge never shares a trunk with other edges, so edges anchored on one side of a node leave or enter it as separate lines, where before two such edges could still run as one; edges without an anchor can still share a line, so anchor the ends you want apart. Two ends anchored closer than 12 px are spread to 12 px apart, evenly over the side when it is too short, in the order of where their other ends lead, so they don't cross. Saved versions keep their stored images.
fixed
Opacity fades the whole element
An element's style.opacity (0 to 1), accepted before but never drawn, fades the whole element as one, as CSS opacity does: a node's shape, icon and label, or an edge's line and its label. A visual diff, from a ?from= image or the Color what changed checkbox on a diagram's dashboard page, mutes what didn't change to half and draws a removed element's ghost at 60%, as the Rendering docs page describes, where before both were drawn at full strength. New saved images include this, and saved versions keep their stored images.
fixed
Label text stays readable on every surface
Label text is checked against every color under it, and only text that falls short changes. A color from the theme, or a white, gray or black the scene sets, becomes near-black or white, whichever reads better, and a color with a hue keeps its hue and changes lightness only as far as it needs to, up to 4.5:1. This covers edge labels, labels drawn outside their shape (BPMN events and gateways, UI flow entry and exit points) and labels on icon nodes with no fill, in every theme, including dark, where many in the playground were unreadable, and saved versions keep their stored images.
fixed
Column, card and header text stays readable
Text that diagram families draw themselves, such as ER columns and types, org card names and titles, UI flow title bars, lifeline headers, notes and fragment guards, is checked for contrast against what it sits on in every render, as labels are. Only text that falls short changes: the muted gray of ER column types and FK badges, and of org card titles, darkens just enough to reach 4.5:1 on white. New saved images include this, and saved versions keep their stored images.
fixed
Gateway and event marks always stay visible
The marks a node draws on its fill (such as a gateway's cross or plus, an event's clock or envelope, or an ER weak entity's inner border) reach 3:1 against the fill in every render, so a gateway's type stays visible even when its stroke matches its fill. A theme stroke that falls short gives way to near-black or white for the marks, an authored one keeps its hue and turns lighter or darker, and the outline keeps the stroke as set. New saved images include this, and saved versions keep their stored images.
fixed
Label backgrounds match every surface under them
An edge label's background takes the color of each surface under it, whether the canvas, a frame or a node, so a label across a frame's border or a node's outline no longer sits on a box in the wrong color. When no one text color reads on every part, the parts it can't read on take the color covering most of the background. A background set with labelBackground stays one color, and saved versions keep their stored images.
fixed
More CSS color forms draw correctly
Colors written as hsl(), oklch(), oklab(), lab(), lch(), hwb(), color(), 4- or 8-digit hex or a var() with a fallback are read correctly in every render, so a dark canvas written that way picks the dark theme and every contrast check sees the color as drawn. PNG images draw them as the SVG does, where some drew black before, including the PNGs of versions saved earlier, whose stored SVGs are otherwise unchanged. Forms the renderer can't read, such as color-mix() or a var() with no fallback, are drawn as written, as the Rendering docs page lists.
fixed
See-through canvases and fills count as they look
A see-through color is read as it shows over what is under it, so a clear canvas such as rgba(0, 0, 0, 0) no longer picks the dark theme and paints a dark canvas, and a label on a clear fill is checked against what shows through instead of drawing white on white. New saved images include this, and saved versions keep their stored images.
fixed
Self-loop labels keep off their loop
A label on an edge that starts and ends on one element, such as a self-message in a sequence diagram, sits above, beside or below its loop, where before, with no clear place beside the loop, it could sit on the loop and its background hid part of it. Saved versions keep their stored images.
fixed
Self-messages and self-references no longer warn
A sequence diagram's self-message and an ER table's self-reference, such as a parent_id column pointing back at its own table, no longer get SELF_LOOP_EDGE_WARNING, since both are drawn as loops on purpose. Every other edge that starts and ends on one element still gets the warning.
fixed
Edges enter boxes clear of their corners
An edge enters a box at least 8 px from the corners of the side it meets, below a UI flow screen's or modal's title bar and clear of a database cylinder's lid seams, so its arrow no longer lands on a corner. Most edges run straight across where the source lines up with that side and enter the middle of the side otherwise, while sequence flows and UI flow transitions enter the middle, as the Rendering docs page describes. New renders and new versions route this way, and saved versions keep their stored images.
fixed
Edges end on gateway and decision diamonds
An edge into a gateway, a decision, an er.relationship or any shape: "diamond" ends on the diamond's outline, not in the empty space of its box. On a wide, flat decision entered from the left or right, the line meets the diamond's point, so the arrowhead stays whole instead of lying along the outline, and the line stays straight where nothing is in the way. Saved versions keep their stored images.
fixed
Straight and curved edges reach their own targets
When two straight or curved edges leave or enter the same side of a box, each ends on its own target and a curved edge stays a curve, where before the arrows of two such edges leaving one side ended in empty space beside their targets and a curve sharing a side could turn into a stepped line. Orthogonal edges route as before, except where the engine spreads the start of an edge into a frame along a shared side: the edge takes a right-angled step, where before its end on the frame moved along too. New renders and new versions route this way, and saved versions keep their stored images.
fixed
Event and gateway labels no longer move shapes
The label drawn outside a BPMN event or gateway, or a UI flow entry or exit point, no longer moves its shape from where you placed it or pushes its neighbors away: it pushes only a node or frame it would come within 8 px of, and only far enough to leave 8 px. An event or gateway drawn on a box no taller than its shape stays centered on that box from its next edit, and a lane or pool with a fixed size doesn't grow for the label, so the event or gateway gets NODE_OUTSIDE_FRAME when its label crosses the lane's or pool's border. Saved versions keep their stored images.
fixed
Workflow: Message flows run straight between overlapping boxes
Workflow: a message flow (workflow.messageFlow) between two boxes whose facing sides overlap but whose centers don't line up runs as one straight line through the middle of the overlap, instead of bending into a hook, wherever that line is clear of other nodes, frame borders and titles. Two message flows running both ways between the same boxes run as two straight lines at least 12 px apart where the overlap has room, and without an overlap a message flow takes a bend. Saved versions keep their stored images.
fixed
Sequence diagram: Message labels sit above their own arrows
Sequence diagram: a message's label sits above its arrow, and a self-message's above its loop, or beside it when the row above is taken, where before messages between the same lifelines could drop a label below its loop or onto its own line. Each row keeps its label's height clear above its arrow, below the message before it, a note, a fragment's header, an operand's guard or the end of a fragment, where labels after a note, at the start of a fragment or operand and after a fragment ended used to drop below or onto their arrows. A label goes below its arrow only when nothing above or beside it is clear, as the Rendering docs page describes, and saved versions keep their stored images.
fixed
Sequence diagram: Fragment operator tabs read in every theme
Sequence diagram: a fragment's operator tab (alt, opt, loop) is readable in every theme, where clean drew white on light gray and the dark theme's tab fell just short. Where white falls short, the tab moves a small step lighter or darker, or the operator turns near-black, and where neither reads the tab moves until the text reaches 4.5:1. Tabs that already read are unchanged, new saved images include this, and saved versions keep their stored images.
fixed
Sequence diagram: Note corners stay visible on dark notes
Sequence diagram: a note's folded corner is its fill a step darker, by at least as much as a light note's, so the fold shows on a dark note, where before it barely did, and a very dark note folds to black. A translucent note's fold keeps the note's transparency instead of drawing an opaque corner. New saved images include this, and saved versions keep their stored images.
fixed
Sequence diagram: Empty fragments no longer draw a stray box
Sequence diagram: a fragment that holds no message or note, such as a ref listing only the lifelines it covers, isn't drawn, with or without a layout of its own, where before it showed as a stray box over the lifeline headers. Its FAMILY_SEQ_EMPTY_FRAGMENT warning says it isn't drawn, so list at least one message or note in its children, a ref's included, or remove it.
fixed
Sign-in fields show the keyboard focus ring
Fields on the sign-in, sign-up, email verification and password reset pages show the keyboard focus ring, where before they showed none. The Google and GitHub buttons work from the moment the page shows, where a click before it finished loading could fail, and work again after you return to the page with the browser's Back button.
fixed
Share no longer opens a picker on iPhones
On an iPhone, opening a diagram's Share dialog no longer opens the Who can see it picker by itself, where before the dialog started on that control and iOS opened its list over the page. Focus starts on the dialog's title, and after you choose Change, focus goes to the sentence that says who can see the diagram.
fixed
Publishing no longer moves a diagram up the list
The Diagrams list's Last changed time and order, and updatedAt in the API and diagram_find, follow a diagram's latest version only. Publishing or unpublishing it, and a team owner's Make all private, no longer move it up the list.
fixed
A lookup by key answers as the list does
GET /v1/diagrams?key= answers the same item that GET /v1/diagrams lists, leaving key out when the diagram has none instead of answering key: null.
fixed
MCP tools show readable names in more clients
Every MCP tool also gives its title in annotations.title, where clients on MCP 2025-03-26 read it, so they show names such as Find diagrams instead of diagram_find.
breaking
Sequence diagram: Fragment operands start at a message or note
Sequence diagram: each operand in a fragment's extensions.operands is { startsAt, guard }, with startsAt the ID of the message or note it starts at, after the fragment's first row, so its divider stays between the same two rows whatever is added above or inside the fragment, and an operand with a y is refused. Diagrams saved before 8.0.0 with y operands still draw, each operand at the first row at or past its y, but a scene from view=dsp that holds them isn't valid input as it is, so replace each with startsAt before sending it back. There is no updateFrame, so to change such a fragment's operands, delete the fragment and create it again with createFrame in one edit, as the Operations docs page shows, and the published JSON Schemas at https://zindex.ai/schemas/ have the new shape.
breaking
Node ports and endpoint portId are removed
Node ports and an endpoint's portId, which never changed a drawing, are removed: the API drops both from every scene and edit it receives and returns a PORTS_REMOVED warning for each node or edge that set them, pointing to the endpoint's anchor instead. The empty ports: [] that versions saved before 7.0.0 carry is dropped with no warning, so a diagram you fetch and send back still works, and an edit saves a version without it. The validation codes PORT_NOT_FOUND, PORTS_NOT_ALLOWED_ON_KIND and PORT_ID_DUPLICATE are gone, and the published JSON Schemas, at https://zindex.ai/schemas/ and in the open-source spec repository tagged v7.0.0, no longer have either field, so check a version saved before 7.0.0 with POST /v1/diagrams/validate, or delete its ports first, as the Schema docs page describes.
breaking
New accounts start with no API key
No sign-up makes an API key, whether by email and password, GitHub or Google, and apps that sign in to Zindex (Claude, Claude Code, Cursor, VS Code and ChatGPT) need none. Until the account's email is verified, creating a key (POST /v1/auth/keys) or allowing an app to connect answers 403 EMAIL_NOT_VERIFIED, as saving a diagram already did, and the dashboard's API keys page and the sign-in page for a connected app show a button to verify the email where Create new key and Allow would be. GitHub and Google sign-ups are verified at once, and a verified account creates keys and connects apps as before.
breaking
Sign-up refuses disposable email addresses
Sign-up by email and password, GitHub or Google refuses disposable inboxes and asks for an address you'll keep, so password resets and team invitations still reach you. A team owner can't list a disposable domain for the team (400 PUBLIC_MAIL_DOMAIN) or invite a disposable address that has no account, since nobody could sign up there to accept (400 EMAIL_NOT_ALLOWED). Existing accounts sign in, reset passwords and accept invitations as before.
breaking
Connected-app endpoints answer keys with SESSION_REQUIRED
GET /v1/oauth/grants, POST /v1/oauth/grants/{id}/move and DELETE /v1/oauth/grants/{id} answer an API key with 403 SESSION_REQUIRED, as the other management endpoints do, where before a key got 401 AUTH_REQUIRED from listing and disconnecting. These routes have only ever served the signed-in dashboard, so only a caller sending a key sees the change, and their 404 NOT_FOUND carries reason app, or workspace when a move names a workspace you aren't in.
breaking
POST /v1/workspaces is for Zindex staff only
POST /v1/workspaces creates a team workspace for Zindex staff only, for the teams Zindex provides, and answers everyone else 402 PAYMENT_REQUIRED, pointing to POST /v1/billing/team-checkout with { name, interval }, which answers { url, seats }, a Stripe Checkout for 2 seats. After paying, POST /v1/billing/team-checkout/complete with { sessionId } makes the workspace if it isn't made yet and switches the session to it, answering 409 CHECKOUT_INCOMPLETE until payment completes, as the Workspaces docs page describes. These routes serve only the signed-in dashboard, so no API key or MCP client could ever call them, and UNPAID_WORKSPACE_EXISTS is gone with the unpaid team workspaces it limited.