New Bring your own storage. Keep your diagrams in your own private GitHub repository, on Pro and Team.Bring your own storage on GitHub, for Pro and Team. Learn more about bring your own storage

changelog

New features, improvements, and fixes to the Zindex platform.

Week of · 112 changes · Every week

new

Diagrams can live in your own GitHub repository

Bring your own storage lets an owner of a workspace on Pro or Team keep its diagrams in a private GitHub repository they control, connected from the Storage page in Settings through the Zindex GitHub App. Every version and its picture becomes a commit there, with the edit's message as its subject. Zindex's own storage stays the default for every workspace on every plan, and nothing moves unless an owner chooses it.

new

A connected workspace's index holds no diagram content

For a connected workspace, Zindex keeps an index of metadata, such as IDs, numbers, times, hashes and settings, including a hash of each diagram key rather than the key, and never the diagrams' content. It fetches each read from the repository, and its database and image store keep no copy except what someone sends with a problem report or feedback, while its edge cache keeps published and shared pages for up to a minute and their images for up to an hour. Zindex detects and reports a history rewritten in the repository, though it can't prevent one, and section 8 of the privacy policy, Bring your own storage, commits Zindex to what it keeps, records and deletes.

new

Agents work unchanged, with two new error codes

Agents see no difference: the same IDs, keys, revision numbers, baseRevision, requestId, hashes and links work in both storages, and no prompt, key or connected app changes. Two error codes come only from a connected workspace. 503 STORAGE_UNAVAILABLE, with retryAfter, means wait and retry, and 409 STORAGE_HOLD, with a reason, means the connection needs a workspace owner, so don't retry.

new

Diagrams move between storages with a progress bar

A workspace that already holds diagrams connects as a move: Zindex copies every version into the repository, checks each against the index's hashes, then switches. Move diagrams back does the same into Zindex's own storage and leaves the repository as it is, on every plan, whether the connection is healthy, on hold or disconnected, as long as GitHub lets the App read the repository. While a move runs, saving pauses and reading keeps working, and the Storage page shows a progress bar with Cancel.

new

Storage settings verify, restore and disconnect a repository

The Storage page in Settings shows the connection and its recent activity, and names the fix when the connection needs an owner. Verify now reads every version back against the index, Restore writes the last known-good files back over a change made outside Zindex, and Disconnect refuses saves until the same repository is reconnected or the diagrams are moved back, while the diagrams stay readable as long as GitHub lets the App read the repository.

new

Bring your own storage adds another account or organization

The repository step on the Storage page in Settings lists the repositories of every installation of the Zindex GitHub App your GitHub user can reach, grouped by account, and offers Install the Zindex app on another account or organization under the field, which opens GitHub's own account picker. A workspace can keep its diagrams in a repository of a company organization even when the App was first installed on a personal account, and the account just installed is listed first. Someone who installs the App from GitHub's own page is told it is installed and continues from Settings.

new

Dashboard pages say when saving is paused

While a connection is on hold or disconnected, or a move runs, the diagrams list and each diagram's page say that saving is paused and that diagrams already saved can still be read while GitHub lets the Zindex app read the repository. Owners get a link to the storage settings, and members and viewers are told that a workspace owner can fix it.

new

Owners are emailed about moves and holds

Every owner of the workspace gets an email when a move of its diagrams, into its repository or back, finishes or stops, and one when the connection goes on hold. The email names the repository, the counts and what to do, never a diagram. Canceling a move sends nothing.

new

A connected repository must stay private

Bring your own storage keeps diagrams only in a private repository, or one internal to an enterprise, so connecting refuses a public one and the Storage page lists it without offering it. A connected repository made public goes on hold, with the STORAGE_HOLD reason repository_public, until it is private again: saving pauses while reading goes on, the owners are emailed, and the notice on the diagrams list, each diagram's page and the Storage page is red rather than amber, since anyone can read the diagrams. A move into a repository stops if the repository goes on hold or is made public, leaving the diagrams in Zindex's own storage and what it copied in the repository, while a hold reported during a move back leaves that move running.

new

Bring your own storage has a docs page

The Bring your own storage page, zindex.ai/docs/storage, is written for a security or compliance review: what the repository holds and what Zindex keeps, what the GitHub App can reach, how anyone checks a history from the repository alone, the limits and holds, and leaving. An announcement bar above the header on the homepage, the docs, the examples and the other public pages introduces the feature and links the page, and it can be dismissed.

new

The official AWS Architecture Icons draw in diagrams

Every aws:* key, the earlier keys saved diagrams use included, draws AWS's own artwork from its Architecture Icons, Release 24, about 780 service, resource, general and group icons, unmodified and at AWS's sizes (a service icon 70 px, a resource or general icon 42 px), with the label below wrapped at 160 px, and icons in a row line up whatever their labels' lengths. An AWS icon draws alone, whatever the node's shape, unless the node's fill is a color other than white, whether its own style.fill or the fill its palette gives it, or the node is an externalSystem, which keeps its dashed box. The new AWS icon keys page in the docs lists every key with AWS's name for it and AWS's attribution, and ICON_NOT_FOUND also suggests icons whose name holds a mistyped key's words, so aws:queue names aws:sqs.

new

Frames draw as AWS's own groups

containerType takes ten AWS group presets: aws.cloud, aws.region, aws.vpc, aws.az, aws.subnet.public, aws.subnet.private, aws.securityGroup, aws.autoScalingGroup, aws.account and aws.corporateDataCenter. Each draws a frame as AWS draws that group, in its color and its solid or dashed line, with no fill, its group icon in the corner and its title at the node label's 14 px, and a frame's own style overrides its preset. The warning AWS_GROUP_UNTYPED names the preset to set on a generic frame titled like an AWS group that holds AWS icons, and AWS_GROUP_NESTING says where a preset belongs when it sits where AWS doesn't nest it.

new

Architecture: AWS diagrams lay themselves out

Architecture: a diagram with AWS Cloud, Region, VPC, Availability Zone or subnet presets needs no coordinates and is laid out as AWS's reference diagrams are, with Availability Zones as equal columns inside the VPC, subnets as rows mirrored across them, an internet gateway or other border gateway docked on the VPC's top border, regional services in a column right of the VPC, global services in a row above the Region and users outside AWS Cloud. The load balancer stands between two AZs, or beside one, only when its edges into the subnets all reach the second row or lower, and otherwise, or with three or more AZs, sits in a row above them, centered. An edit grows rows and columns instead of laying the diagram out again (a node added to a subnet takes a free place in its row, and an AZ added with createFrame's frame becomes a new column, what lies past it moving right), and the layout engine docs give the rest of the behavior: AZs or subnets a Region lists directly are arranged as a VPC with no frame drawn, and security groups, Auto Scaling groups, accounts, corporate data centers and any AZ or subnet outside every Region and VPC are laid out as any frame.

new

Frames take an icon and their own dash

A frame takes an icon, drawn in its header: an aws:* key as AWS draws a group icon, 35 px in the top-left corner with the title beside it (on a frame with no preset and no style.stroke, the border takes the icon's category color, solid), and any other icon 16 px before the title. A frame drawn the generic way follows its style.dash, where before every one was dashed [4, 4]: left out it stays [4, 4], [] draws it solid and any other array is the pattern. Each dashed side is stretched a little to start and end on a dash, so every corner is drawn and a side an edge crosses keeps the same rhythm.

new

Edits put new frames inside saved frames

createFrame takes an optional frame, the ID of a frame that exists when the op runs, and adds the new frame to it as its last child, so one op adds a lane after a saved pool's other lanes or an Availability Zone to a saved VPC. A frame naming an ID that doesn't exist yet is refused with OP_TARGET_NOT_FOUND, and one naming something that isn't a frame with OP_TARGET_KIND_MISMATCH, as for createNode.

new

updateFrame changes a saved frame in place

The new op updateFrame changes a saved frame's title, style, icon, containerType and laneDirection, and null removes any of them but the title, so renaming a frame or making it an AWS group is one op, where before it meant deleting and recreating the frame, which also deleted the edges that end on it. The frame keeps its members, position and size, and its nodes stay where they are, so to have a saved diagram arranged as AWS draws it, send the updateFrame ops with an autoLayout in the same edit. It doesn't change children, which createNode, updateNode and createFrame with frame do.

new

AWS diagrams draw AWS's open arrowhead

In a diagram with an aws.* group, an edge that sets no endArrow and whose edgeType is unset, default, or an er.*, network.* or org.* type ends in AWS's open V (openTriangle) instead of the filled triangle. Workflow, sequence and UI flow edge types keep their own arrowheads, so workflow.sequenceFlow, sequence.message, sequence.create, sequence.destroy and the uiflow.* types still draw the filled triangle. An endArrow the edge sets still wins, so openTriangle draws the V on any edge and triangle the filled one, and the default isn't stored, so the scene shows only what its author wrote.

new

Numbered step badges mark a request's path

An edge's step, a whole number from 1 to 99, draws AWS's numbered callout beside the start of its line, a 25 px circle in the theme's text color where that reads at 4.5:1, else near-black or white, with its number in the canvas color, so it inverts on a dark canvas, and edges with the same step and source share one badge. Badges are placed after the routes, 10 px clear of icons, captions, labels, lines, frame borders and other badges, or where they overlap least when no place is clean; when the engine lays out every node it keeps room for each badge beside its edge's label, while a step added to a saved diagram moves no node. The engine draws the numbers you give without numbering edges or checking their order, sequence diagrams and hidden edges draw no badge (a step in a sequence diagram gets the warning STEP_NOT_DRAWN), and a Mermaid view leaves steps out while a Mermaid edit keeps them.

new

Step text builds a numbered list beside the drawing

An edge's stepText, which needs a step, is its line in a step list 40 px right of the drawing, top-aligned and with no panel, and the canvas grows to hold it with the drawing's margin past it. Steps appear in number order, each with its badge and then its edges' text in document order, wrapped at 280 px, with a bullet for each edge's line when a step has several, and updateEdge changes either field, step: null removing both. The list shows stepText only, so labels stay on their lines, a step whose edges have no stepText has no entry, and a diagram with none has no list.

new

Architecture: Spanning frames draw groups across Availability Zones

Architecture: in architecture diagrams and diagrams without a family, a frame written with spans instead of children is drawn around the nodes and frames it names while each keeps its own frame, as AWS draws an Auto Scaling group or a security group across Availability Zones, with no fill and its border breaking where lines cross it. createFrame takes spans and updateFrame replaces them, deleting an element drops it from every spans list and deletes a spanning frame left spanning nothing, with its edges (SPAN_REMOVED), and since the box is one rectangle, the warning SPAN_COVERS_NODE names a node inside it that it doesn't span and that isn't inside one of its targets. A spanning frame holds no members, sits in no frame or group and spans no other spanning frame (SPAN_INVALID), other families draw nothing and get the warning SPAN_NOT_DRAWN, and a Mermaid view leaves spanning frames out while a Mermaid edit keeps them.

new

Architecture: Spanning frames get room for their headers

Architecture: in AWS diagrams the layout arranges, a spanning frame around subnets deepens the row lane above their top row for its header in every Availability Zone, so its top border runs between the rows, and one around what subnets hold deepens the lane under the subnets' headers instead, an edit that adds one moving what lies below and listing it in changed. The header sits where its preset puts it, an Auto Scaling group's centered under its icon, and moves along the top edge to keep 16 px off lines and other frames' sides, and on a side an arrow comes in by, the border keeps the arrowhead's reach from its end, so it runs through no arrowhead. That room is kept only when every target is a subnet in one VPC's rows or sits in one, so one around Availability Zones or VPCs, or reaching a Region's column, AWS Cloud's row or the outside columns, gets none, and in other layouts each target keeps a margin while the engine places nodes, but one added to a saved diagram moves no node.

new

An example draws AWS architecture from Terraform

The new worked example, AWS architecture diagram from Terraform (aws-architecture-from-terraform), has an agent draw a team's AWS architecture from its Terraform configuration with no coordinates, laid out as AWS's reference diagrams are. An infrastructure pull request that adds an SQS queue and a Lambda worker is recorded as revision 2, with the before and after pictures reviewers see before terraform apply, the prompt, the operations and the example's own preview card.

new

Flowcharts are their own diagram family

diagramFamily: "flowchart" is for steps with yes and no decisions and loops, such as a deploy pipeline, a retry policy or a request handler's checks, while BPMN processes with events, gateways and message flows stay in workflow. A flowchart runs top to bottom unless layoutStrategy.direction says otherwise, aims for 50 px between ranks rather than 80, and gives every line into a step its own arrowhead, joining only lines into one side of a decision, at its point, while edges with the same label each keep it. In the default light style its nodes draw light blue with a blue border, except external systems, notes and nodes with an aws:* icon, while a node given its own fill or stroke keeps it and the dark, blueprint and sketch themes keep their own colors.

new

Flowchart: Yes runs straight on, no turns aside

Flowchart: a decision's exit whose whole label is a yes answer (yes, true, ok, pass and the like) runs straight on and one whose label is a no answer turns right, or down when the flow runs left to right or right to left, so the main path from the start runs in one straight line. When one of a decision's two exits loops back to an earlier step, the other runs straight on whatever its label, and loops back across two or more ranks run round the left, or above in a left-to-right or right-to-left flow, away from the no branches, unless the other side crosses fewer nodes or lines. Decisions whose exits carry other labels, such as approved, keep the general decision rules, an else-if ladder draws as a column of checks, and in a flowchart with a frame that has no layout the main path isn't lined up and no branches aren't turned aside, though its loops still keep their side.

new

Flowchart: Edits insert steps on the main line

Flowchart: a node an edit inserts between two placed nodes goes on the line of the one before it, and every node past it moves along the flow just far enough to make room, while the nodes before it stay and a step that replaces a deleted one takes its place. A new no branch goes beside its decision, to the right, or below in a left-to-right or right-to-left flow, moving nothing but what its label needs. In a flowchart with a frame that has no layout, inserted nodes are placed by the general rules.

new

Flowchart: Two warnings check each decision's exits

Flowchart: FAMILY_FLOWCHART_DECISION_FEW_EXITS names a decision with fewer than two exits, to be given an exit for each answer or made a process, and FAMILY_FLOWCHART_EXIT_UNLABELED names an exit of a decision with two or more exits that has no label, to be given the answer it stands for with updateEdge. Both are warnings, so the diagram still saves.

new

Document and subroutine shapes join every family

Two new shapes work in every family: document, a box with a wavy bottom and its label above the wave, and subroutine, a predefined process drawn as a box with a line just inside each side. A document node draws the document shape by default, where it drew a plain box, and diagrams saved earlier keep the shape they were saved with. Mermaid's document (doc or document) and subroutine ([[ ]], subroutine, fr-rect or subproc) shapes import as these shapes and read back as @{ shape: doc } and [[ ]], where before both were drawn as rectangles and a subroutine was reported with SHAPE_APPROXIMATED.

new

An example draws a deploy pipeline flowchart

The new worked example, Deploy pipeline flowchart from a GitHub Actions workflow (deploy-pipeline-from-github-actions), has an agent draw how a push reaches production from the repository's workflow file, with no coordinates, and record a pull request that adds a canary rollout loop as revision 2, with its before and after pictures and its compare. The homepage's What agents draw has a tenth tab, Flowchart, third after the two architecture tabs, which draws this example, and the playground's Mermaid sample is a webhook retry flow that imports as a flowchart.

new

A new homepage replays agents sharing one diagram

The homepage opens on an example session, replayed: five teammates direct their agents in Claude, Cursor, Copilot in VS Code, ChatGPT and Claude Code, and each agent's diagram_create or diagram_edit call saves the next revision of one architecture diagram, shown zoomed in on the change and colored by what changed. One agent's stale edit is refused with REVISION_CONFLICT, saving nothing, until it reads the newer revision and plans again, and a ring marks each person and a square each agent, so who asked and which agent made each change read at a glance. The chain under the picture opens any saved revision, the replay holds while the pointer rests on it or with Pause and starts paused under reduced motion, and Present opens the picture full screen.

new

The homepage shows what agents draw

What agents draw, on the homepage, has a tab for each of nine kinds of diagram: architecture, AWS architecture, ER diagram, sequence, BPMN workflow, swimlanes, data flow, org chart and UI flow, each with a caption, its diagramFamily and what the agent made it from. Each picture is drawn afresh in Zindex's dark theme: six show a worked example's scene, among them the AWS architecture an agent makes from Terraform, and swimlanes, org chart and UI flow show scenes drawn for the homepage. A tab is chosen by a click, the arrow keys or a resting pointer and loads its picture only then, nothing switches by itself, and Present opens the picture full screen with zoom.

new

The homepage ledger lists every save and refusal

The homepage's Teams section draws five people, the agents they direct and the one diagram every agent saves to, with the stale edit refused, and says a shared workspace takes the Team plan. Its revision ledger lists every saved revision newest first, with who asked, through which agent, its message, what it added, changed and removed, its date and its hash, and the refused edit as its own row in the API's words. A slider steps through the revisions' pictures, Color what changed shows each one colored since the one before, and Present opens it full screen.

new

The homepage links examples, setup steps and pricing

Six use cases on the homepage each open a worked example's page, with its prompt, its operations and the pictures Zindex drew, and show that example drawn afresh in Zindex's dark theme: its scene, or for "A change you can review" its edit, colored by what it adds. Connect has the MCP server's address, a line to paste into your agent, the Claude Code command and each app's step, for Claude, Claude Code, Cursor, VS Code, ChatGPT and an API key, with Copy buttons. Pricing returns as a strip under the price line, "Free to start. $5 a month for unlimited diagrams.", with the Free, Pro and Team plans in brief and a link to the pricing page.

new

The Connect page has a line for agents

zindex.ai/connect has a line to paste into your agent, with a Copy button, that has it connect to Zindex's MCP server, https://api.zindex.ai/mcp, and read the Zindex skill, https://zindex.ai/skill/SKILL.md, first. In Claude Code, which runs commands, your agent can add the server from the line, and you then run /mcp to sign in, in a new session if zindex isn't listed yet; in the other apps you add the server with the app's own step, and the line has your agent read the skill. The page's Copy buttons, like the homepage's, also tell a screen reader when the text is copied, or when to select and copy it.

new

Logs record IP addresses and diagram readers

Request logs record the IP address each API request comes from (cleared after 90 days) and the diagram, key or link it names, and the audit log records which account read a diagram in its workspace. Request logs still record views of public diagrams and share links without the IP address or the diagram. The privacy policy, updated 2026-11-02, lists the change in its Changes section.

new

Billing history opens invoices after a subscription ends

Once a workspace's subscription ends, a Billing history button on its Billing page opens its past invoices. In a team workspace, only owners can open it. Closing an account or deleting a team workspace says to download invoices first whenever there is billing history, not only while a subscription is live.

improved

Team is a plan, no longer a beta

Team is a plan like Free and Pro: the Beta badge is gone from the pricing page, the homepage, a team workspace's Billing page and the page that creates a team workspace, and so is the note that Team's features may change during the beta. The docs, the Zindex skill, the pages written for agents and the OpenAPI description no longer call it a beta, and its price and limits stay as they were.

improved

The pricing page names bring your own storage

The Pro and Team cards on the pricing page and on the homepage's pricing strip list bring your own storage, its name linking the storage docs on the pricing page, and Pro's tagline reads "Unlimited diagrams, in Zindex's own storage or yours." A new first question in the pricing FAQ, "How does bring your own storage work?", says that every workspace starts in Zindex's own storage, that a workspace owner on Pro or Team can connect a private GitHub repository at no extra cost, that Zindex's index holds metadata and never the diagrams' content, and that saves and reads wait while GitHub doesn't answer or is past its limits. The answer on leaving a paid plan adds that a connection keeps working within Free's limits, and the plan cards line up their dividers, feature lists and buttons whatever the length of each card's tagline.

improved

A move shows its steps and starts at once

While a workspace's diagrams move, the Storage page opens with the move: its four steps, Copying, Verifying, Switching and Finishing, with the current one marked, the phase and count as its main line, a progress bar across the card, when the move started and last progressed, and Cancel. The first batch starts within seconds of connecting a repository or choosing Move diagrams back, and after a move fails or is canceled the page offers Try again. When nothing has changed for thirty seconds, a line says the move is still working.

improved

A move says why it waits for GitHub

When GitHub doesn't answer, the move pauses and the Storage page says it is waiting, why and since when; nothing is lost, and the move goes on from where it stopped once GitHub answers, in either direction. Move diagrams back switches the workspace to Zindex's own storage in one step, so an interruption at the switch can't leave it half moved.

improved

Storage settings show the time of day

The Storage page in Settings gives the time of day with every date, in your own time zone, so you can tell at a glance how far apart events were. Recent activity marks today's rows with a green Today badge, as the Diagrams list does.

improved

Regular storage checks share one line

On the Storage page, the regular checks that find every sampled version matching share one line at the top of Recent activity, such as "Checked 4 times since 20:30; all match", so a run of them no longer pushes a hold, a mismatch or an owner's action out of the list. An owner's Verify now still gets a row of its own, and so does any check that finds a mismatch.

improved

Chains of nodes line up, with fewer bends

In architecture, workflow and UI flow diagrams and diagrams without a family, laid out without frames, and inside an AWS diagram's frame placed as one block that holds no frame, such as a serverless diagram's AWS Cloud or Region, each chain of nodes lines up on one straight line, long edges taking priority, so across Zindex's 131 test diagrams bends fall from 1,289 to 1,220 with no new crossings. Where two or more of a decision's or gateway's branches leave one side, each target ends on its line or sits clear of it, when every such move in the diagram keeps the layout's spacing, so a branch runs straight or leaves its own tip and turns once, and a drawing whose decisions branch both ways can come out wider. Fresh layouts, autoLayout and nodes an edit places take the new placement while saved diagrams keep their positions, and entity-relationship diagrams and diagrams whose frames are laid out with their members, pools and lanes included, keep their placement.

improved

Long edges reroute around the lines they cross

In workflow, UI flow and architecture diagrams and diagrams without a family, but not AWS diagrams the layout arranges, an edge that runs with the flow across two or more ranks and crosses another line is routed again, and may leave or enter by another free side with up to two more bends, one for each line it no longer crosses. The new route is kept only when the finished diagram has fewer crossings and is no worse by any other measure, labels on other edges' lines included. At most eight reroutes are tried per layout, and a diagram of more than 250 edges keeps its first routes.

improved

Architecture: AWS gateways dock on the side they face

Architecture: an AWS gateway, one an edit adds included, docks on the VPC's border facing what it connects to outside. A gateway to users on the left docks on the left, but one whose lines reach several AZs still docks on top. A load balancer in a VPC with one Availability Zone stands on the side its traffic comes from.

improved

Architecture: AWS layout reads through wrapping frames

Architecture: a generic frame around all of a VPC's Availability Zones, such as an EKS cluster, or around whole VPCs keeps AWS's arrangement inside it. Any other generic frame between AWS groups loses it, and the new warning AWS_FRAME_HIDES_GROUPS names the groups it hides. The warning is raised only in architecture diagrams and diagrams without a family.

improved

Architecture: AWS layout reads through tier frames

Architecture: a generic frame in an Availability Zone that holds only subnets next to each other in its rows, such as a tier, keeps AWS's arrangement: its subnets join the AZ's rows, shared with the other AZs, and the frame is fitted 10 px around them, its header in room kept above its first subnet. A diagram saved with such a frame drawn as its own box keeps its nodes where they are until autoLayout, and a tier across several AZs is one frame with spans naming their subnets. A frame around subnets with another between them, or around subnets and nodes, still keeps its own box, with the warning AWS_FRAME_HIDES_GROUPS, whose fix names spans when the frame hides subnets.

improved

Architecture: AWS lines keep off group titles and borders

Architecture: in AWS diagrams the layout arranges, lines keep clear of every group's title, an Availability Zone's too, where a clear route exists. A line running along a group's border moves off it, and two edges' last legs running side by side are nudged apart. A move is kept only when it adds no crossing.

improved

Architecture: AWS layout makes room for labels and lines

Architecture: in AWS diagrams the layout arranges, two nodes in one subnet joined by a labeled line keep room between them for the label, clear of any fork on that line, and a load balancer between two Availability Zones keeps at least 64 px each side, so its lines into the AZs run 32 px or more from the AZ borders and from its own box. The column of regional services stands far enough right that a labeled line into it, such as "logs, metrics" into CloudWatch, keeps a clear stretch past the lines crossing it, on a fresh layout and after an edit that places nodes in the column or joined to it, an order constraint added there included, where the whole column, and what lies right of the Region, moves right as one. A node outside AWS Cloud whose line runs into a gateway on a VPC's top or bottom stands where that line passes 16 px under every group header it crosses.

improved

Edges meet AWS icons, not their labels

In every family, lines ending on one side of an AWS icon keep 16 px apart where it has room. An end on a round icon, such as a gateway or load balancer, meets its circle. An arrowhead from below meets the icon's side rather than the label under it when a route one bend longer exists.

improved

Edge labels slide clear of frame borders

In every family, a label slides along its line off a frame's border where there is room. In AWS diagrams the layout arranges, a label also doesn't sit in line with another edge, so it reads as its own line's.

improved

AWS's open arrowhead also applies without AWS groups

AWS's open V (openTriangle) is also the default arrowhead in diagrams with AWS icons and no aws.* group. Workflow, sequence and UI flow edge types keep their own arrowheads, and an endArrow the edge sets still wins. In every family, a dashed line into an open V ends on a dash at its tip.

improved

Large database schemas lay out several times faster

An entity-relationship schema of 80 tables and 231 references, laid out in full with the engine comparing its candidate layouts, takes 7.2 s where it took 34.4 s, past the 20-second drawing budget, so it answered DRAW_TIMEOUT, and one of 160 tables takes 13 s where it took 90 s, as measured by Zindex. The passes that got faster belong to the edge router every diagram family shares, so other diagrams with many edges can gain too, while small diagrams take about as long as before. The speed-up changes no drawing: the faster passes choose exactly what the slower ones did.

improved

Dots mark forks, and bundles keep separate lines

In architecture, ER and network diagrams and diagrams without a family, a dot marks every point where edges drawn alike fork from a shared line, a T or a four-way fork, while lines that only cross stay plain, so a fork no longer reads as a crossing; workflow, UI flow, org chart and sequence diagrams draw no fork dots. A fan-out's bus runs 24 px from the faces it feeds, or midway in a narrower gap, so its dots sit out in the channel, branches keep 9 px apart where there is room and crowded dots draw smaller, so no two dots touch. Different bundles that would run end to end on one line less than 32 px apart take separate tracks up to 24 px apart, so two buses no longer read as one, and in a colored diff a dot takes the status of the lines through it.

improved

Colored diffs start a new branch at its fork

In a colored diff, the dot where an added or removed branch leaves a line both versions draw is drawn in that branch's color, green for added and red for removed, so a new line visibly starts at its fork. Where an added branch and a removed one part at one dot, the dot is the added one's, and a dot an unchanged branch leaves from keeps the status of the lines through it. The stretch of line both versions draw stays unchanged, in its own faded styling, as a text diff's context lines do.

improved

Colored diffs show step badges and spanning frames

In a colored diff, a step badge takes the status of the edges it stands for, drawn changed when they differ, with a 3 px ring in the added or changed color or a dashed ring on a removed ghost, faded as its edges are. A badge in the step list does the same for the edges whose lines follow it, and each line fades or ghosts with its edge. A spanning frame is drawn around its targets in both versions, so a removed target's ghost sits inside it, and its border takes its status as a frame's does, so a target the comparison adds or removes doesn't make it amber.

improved

Diagrams without a family draw as architecture

A diagram that declares no diagramFamily is drawn as an architecture diagram, so edges with the same label into one node each keep their own label, end and arrowhead, where before they could join into one labeled line, and extensions.autoGroup draws its bands. A diagram saved without a family keeps its node positions until it is laid out again, and declaring flowchart gives it a flowchart's conventions.

improved

ER diagram: Crossing relationships can go around a table

ER diagram: a relationship that ends on a table's row and crosses another line can be routed around a table instead. It goes around only when that path takes no more bends, stays inside the drawing and leaves fewer crossings, and its line can run longer. On the homepage's commerce schema, orders.customer_id runs above addresses instead of crossing orders.address_id beside a crow's foot. Schemas of more than 250 relationships aren't rerouted, and smaller ones whose relationships cross lay out more slowly (about 1.5 s at 40 tables, from 0.9 s, as measured by Zindex).

improved

"character" wrap breaks long paths at slashes

With textStyle.wrap: "character", a word wider than its line breaks after a / first, then after a -, and between characters only where a piece is still too wide, so s3://acme-media-uploads/originals/2026/10/07/ reads s3://acme-media-, uploads/originals/ and 2026/10/07/ on three lines.

improved

The homepage replay shows one moment at a time

The homepage's example session plays one moment at a time: the five people and their agents sit above the picture as chips, the one at work lit; the agent's pointer and name rest on the part of the diagram it is about to change; then the saved revision appears colored by what changed, the rest of the picture dimmed, with one sentence under the picture saying what happened over the call's receipt (its operation count, the counts added, changed and removed, and the hash). The one refused stale edit shows as a red pointer with the API's 409 status and a struck-out mark where the stale change would have gone. The card opens on the finished diagram, each person's tag parked on the element their agent changed and the revisions below filling in as they are saved; step buttons, Pause, the chips and the revision rail all move through the session, under reduced motion it stays on the finished diagram, and the loop runs about 32 seconds instead of 63.

improved

Example pictures open at full size

On each example page, Open full size under the rendered diagram, and under each of the Revision diff's before and after pictures, opens its SVG in a new tab. The before and after pair stacks at every width, so each picture draws about twice as large as it did side by side. Every example's pictures, preview cards and diffs, and the homepage's examples, are redrawn with this release's engine.

improved

Pasting a verification code submits it

On the email verification page, pasting a code of exactly six digits fills the boxes and submits it at once, ignoring spaces, dashes and anything else that isn't a digit. A shorter paste fills the boxes it can, and a longer one fills them with its first six digits and waits for Verify, since each try uses one of the code's five attempts.

improved

Checkout's refusal links to email verification

When checkout on the pricing or Billing page is refused because the account's email isn't verified, the message ends with Verify it now, which emails a new code, opens the verification page and then brings you back to the page you were on. Before, the message gave the verification page's address as plain text. The banner asking you to verify names buying a plan too, alongside saving diagrams, creating API keys and connecting apps.

improved

Verifying from the pricing page reopens checkout

When Start Pro on the pricing page is refused because your email isn't verified, Verify it now brings you back to Pro checkout once you verify, monthly or yearly as you chose, where before you pressed Start Pro again. The pricing page also shows the banner asking you to verify your email, so a signed-in account that hasn't verified learns it before choosing a plan rather than after.

improved

The changelog has a page for each week

The changelog at zindex.ai/changelog/ shows the newest week's changes, then a list of every week with its number of changes. Each week has a page of its own at its Saturday, such as zindex.ai/changelog/2026-10-31/, with Newer and Older links to the weeks either side, and its address stays the same as newer weeks arrive. A page holds one week rather than every change since April, so it loads far less than the single page did.

improved

Link previews carry the new short description

The homepage's description, which link previews show, reads "The diagram runtime your agents share. Zindex validates each saved change and keeps its revision and image.", and zindex.ai pages without a description of their own give the same line after "Immutable agent diagrams." A shared diagram's page says the same in its box for visitors new to Zindex.

improved

Shorter docs leave prices to the pricing page

The Workspaces page is rewritten in plain words at under half its old length, and names an error code only where an agent receives one. No docs page gives a price: the price row of the Limits page's plans table links the pricing page instead. The docs and the OpenAPI spec list only the requests an API key or a connected app can make, and point to the signed-in dashboard for the rest, such as moving a diagram into a team workspace.

fixed

Connecting finds the App's existing installations

The GitHub step starts with GitHub's sign-in and lists the repositories of every installation of the Zindex GitHub App you can reach, so connecting works when the App is already installed on your account or organization. GitHub's install page opens by itself only when there is no installation yet.

fixed

A move with large pictures no longer stalls

A move of a workspace's diagrams into its own repository copies versions in commits sized by their files, about 8 MB of pictures each, so a workspace with large diagrams goes in as many commits as it needs instead of stalling partway with no progress. A move that stalled this way can be started again from the Storage page in Settings, and it resumes from what the repository already holds.

fixed

Pruned histories and deleted diagrams move too

A move, and Verify now, take a diagram's history from its first kept version, so a diagram whose oldest versions an earlier retention limit removed moves and verifies like any other. A diagram deleted in the last 24 hours, which can still be restored, moves with every version and its pictures. A move that fails names the reason on the Storage page.

fixed

Repository records match their hashes, frames included

Each version's record in a connected repository holds the scene as written, as a read returns it, so the SHA-256 of its RFC 8785 form matches the record's content hash for every diagram, frames included, and a history checks out from the repository alone. Versions saved earlier keep the files they were saved with, and Zindex reads, verifies and moves them as before.

fixed

Colored diffs paint only what an edit changed

A colored diff no longer draws an element amber when its only change is its position or size, as when an edit grows a row and the engine moves what lies past it, nor a frame whose only change is members the same edit adds or removes, which are drawn green or red already, while a change to a label, type, shape, style, icon, frame membership, edge end or cardinality stays amber, moved or not. The compare still lists all of them as changed, and a position or size that moves by 0.01 px or less is no change at all. A changed node drawn as its AWS icon alone is outlined around its icon and caption, and no edge label covers the outline.

fixed

Colored diffs draw removed members inside their frames

In a colored diff, a removed member's ghost is drawn back inside the frame that held it, where before it could be laid out on its own outside the frame, such as a removed DynamoDB table drawn beside its AWS Region instead of in the Region's column.

fixed

Colored diffs ring frames in their status color

In a colored diff, an added, changed or removed frame keeps its own border, with its color, width and dash, and its status is a ring 4 px outside it, 3 px and solid in the added or changed color or 2 px and dashed around a removed frame's ghost, breaking around an icon on the border, while a lane's ring runs 4 px inside its border. Before, the border itself was redrawn in the status color, so it no longer said what kind of group the frame was. The comparison keeps room for each ring: an edge label keeps 6 px from it where its edge has a clear place, a sequence diagram's rows after a ringed fragment move down to keep their gaps, and the frames around a ringed spanning frame grow in place, while lines still cross a ring and the nodes each version placed keep their places.

fixed

Colored diffs hide the line under a faded label

In a colored diff, an unchanged edge's label, or a removed edge's ghost label, keeps an opaque background in the colors of what lies under it and fades only its text, no further than 4.5:1 on that background. Before, the background faded with the text, so the line the label covers showed through it.

fixed

Colored diffs strike through a removed step's text

In a colored diff, a removed edge's lines in the step list and their bullets are drawn in the removed color and struck through, so the removal doesn't rest on color alone, faded as the edge's ghost is but no further than 4.5:1 on the canvas. Before, they were drawn in the text color, only a little darker than the unchanged lines, so the step read as still taken.

fixed

New frame members land clear of other frames

A node an edit adds to a placed frame goes clear of every frame it isn't in, where before, with no room by its neighbors outside other frames, it could land inside an unrelated one, such as a new queue inside an Availability Zone. When its own frame has a layout with no room left, it goes just outside that frame, still clear of the others, and NODE_OUTSIDE_FRAME says so.

fixed

Edges route around frame titles and icons

A frame's header, its title and any icon, is an obstacle for every route that can reach its end another way, so edges no longer cross a group's title when a clear route exists, and keep 16 px from a header that carries an icon. Only when no route keeps off every header does an edge cross one.

fixed

Edge labels keep clear of other nodes

An edge label keeps 6 px from every node but its own edge's ends, so it no longer reads as another node's caption, and labels in workflow, org chart and architecture diagrams that sat closer move. Two edges that cross in a tight channel trade tracks when that removes the crossing. A label with too little of its line left beside a frame's border sits beside its own bend rather than over its line.

fixed

Edge labels no longer cover frame borders

In every family, an edge label tucked beside a frame's border no longer erases part of it. Where only the label's padding would reach a border its text doesn't cross, the background behind the label stops half the border's stroke and a pixel short of it, so the border is drawn whole. The label's text and place don't move, and a border the text does cross stays hidden under the label as before.

fixed

Edge labels keep away from other lines' ends

In every family but sequence diagrams, an edge label no longer sits beside a place where another line crosses its line within two label heights of the other line's arrowhead or ER glyph, where the label read as the other line's. The label is centered on the longest stretch of its own line between such crossings, and a line that only crosses, its ends farther off, doesn't move it. On a short run with several such crossings, the label can still sit near one.

fixed

Step badges no longer face each other across lines

A numbered step badge no longer stands opposite another badge across that badge's line, where the two read as a pair and the one whose line it isn't read as that line's. Once every badge is placed, a badge within 62.5 px of another badge's line, opposite that badge, moves once to the first clean place along its own line where it faces no badge and none faces it. Where its line has no such place, it stays.

fixed

Edges leaving one side no longer overlap

Edges leaving one side of a node no longer run one over another on one line, or end to end less than 32 px apart: each keeps its own track. Labeled exits that turn the same way step out as a staircase, each turning past the one before by at least its label's width plus 12 px either side, so each label sits beside its own line, where that crosses no more lines and keeps 16 px from other nodes. AWS diagrams the layout arranges keep their own spacing for lines end to end and draw no staircase.

fixed

Edges ending side by side keep separate arrowheads

In workflow, UI flow and org chart diagrams, which draw no fork dots, edges ending on one side of a node each keep their own point and arrowhead instead of joining another at a dot, spaced closer than 12 px where the side is short. Entries into one side of a flowchart decision (not a BPMN gateway or workflow.decision, which keep one arrowhead per flow) and edges whose shared label is drawn once still join at a dot, as ends sharing a point do in architecture, entity-relationship and network diagrams and diagrams without a family. In every family, two parallel edges running straight between the same two nodes are drawn as two lines, even beside a fan-out's shared trunk.

fixed

Lines meeting one side from one direction nest

Two lines that run along one side of a node from the same direction, turning into it or leaving it, nest: the outer line takes the outer end, so neither crosses the other on its way in or out, where before the two could take their ends in the wrong order and cross just short of the node. Where their routes cross, one line's run moves a track over, kept only when the drawing then has fewer crossings and neither line runs through a node or onto another line. An end whose place is fixed, an anchor with an offset, an ER row's end or a bundle's shared trunk, doesn't move.

fixed

Parallelograms and hexagons fit their labels

A parallelogram leans by half its height and a hexagon's points reach a quarter of its height past its top's corners, each at most a quarter of its width, and both widen so the label fits between the slanted sides, where before it could run past them. Edges into and out of a parallelogram's top and bottom run straight instead of jogging, and an edge leaving a parallelogram's or a bucket's side in a left-to-right flow leaves by that side rather than going round by another.

fixed

Arrowheads stay off rounded corners and pill ends

An arrowhead or ER glyph on a node with rounded corners keeps 6 px past the corner's curve, so its wings stay beside the straight side: 14 px from a roundedRect's default corners, more with a larger cornerRadius, and half the height plus 6 px from a pill's ends, which take an end only at their middle. Before, an arrowhead could land on a pill's rounded end, and ends spread along a rounded box's side could come closer to its corners. Where the straight part has no room for a straight line, the edge bends into the side's center instead.

fixed

Back loops keep their full clearance from nodes

An edge that loops back against the flow keeps its full 16 px from the nodes it passes, where before it was drawn 1 px nearer and could raise EDGE_OBSTACLE_GRAZED.

fixed

Architecture: Order constraints hold in AWS rows and columns

Architecture: an order constraint an edit adds to a saved AWS diagram puts its two nodes in the order asked within their row or column, the two trading places when they stand the wrong way around, and nothing else moves, where before the pair stayed the wrong way around and other nodes could move. Nodes an edit adds and joins by an order constraint in a subnet's row, AWS Cloud's row, a Region's column or the column of items outside AWS Cloud are placed in the order asked. A Region's column that splits in two on a fresh layout breaks where no above or below pair straddles the split, where before such a pair could stand the wrong way around across it.

fixed

Architecture: AWS lines keep clear of icons and captions

Architecture: in AWS diagrams, two lines meeting an icon's opposite sides level with each other no longer read as one line straight through it, since one moves along its side, a straight line between two icons moving whole, and a fan-out's lines keep out of the caption of the icon they leave. An arrow from below into an icon lying well to one side also comes in by its side, not under its label. A lone line in a narrow gap between groups runs down the gap's middle, and lines spread across such a gap keep off an Availability Zone's title too.

fixed

Architecture: AWS lines pass row-mates in the row's lane

Architecture: in AWS diagrams the layout arranges, a line from a subnet's member that passes a row-mate on its way to the Region's column, or to an item beside the VPC, runs in the lane under the row, as a line between row-mates already does, where before it could cross its neighbors' lines just above them. It takes the lane only where the lane and the way on are clear, and otherwise keeps its route.

fixed

Architecture: Lines beside a lone AZ keep 32 px

Architecture: an item beside a VPC's only Availability Zone, such as a load balancer, keeps its lines 32 px or more off the AZ's border, as an item between two AZs does, where before they ran 20 px outside it and read as one more border. A line from it into a member behind a nearer row-mate comes down into the member's top from the row's lane. The wider spacing comes with a fresh layout or autoLayout, and a saved diagram keeps its positions until then.

fixed

Architecture: AWS lines that read as one move apart

Architecture: in AWS diagrams the layout arranges, when an end leg that can't move, such as a bundle's shared trunk, an end on a face its family fixes or an end with an anchor, runs nearly in line with another line's middle segment, that segment moves across to 16 px off it, with its bends, unless another line's run shares it, so the two no longer read as one line. An end leg nudged off another line can also move past it to 16 px on its far side when the drawing then has fewer crossings, where before it always moved to the side it lay on, even when that kept it crossing the other line. Each move is kept only when it adds no crossing, keeps off every node and leaves fewer such pairs.

fixed

Architecture: AWS lines no longer merge or run through icons

Architecture: in AWS diagrams the layout arranges, a line's last leg moved apart from another line is never laid on any other edge's line, where before a line leaving a node could be drawn on the line coming into the same side, as one line that forks. Nor does the move leave two lines level on an icon's opposite sides, reading as one line straight through the icon. A move that would do either is refused, and the line stays as routed.

fixed

Architecture: AWS tiers nest 20 px inside their zone

Architecture: in AWS diagrams the layout arranges, a tier, a generic frame around an Availability Zone's subnets, stands 20 px out from them, and the zone's border 20 px out from the tier on the left, the right and the bottom, as AWS nests one group in another, where before the tier ran 10 px inside the zone's border and read as a double border. Every Availability Zone of a VPC that holds a tier starts its subnets 40 px in, so every subnet keeps one width and the zones stay as wide as each other. A spanning frame still keeps 10 px around its targets. In a saved diagram, whether an edit adds the tier or the diagram was saved with one, what lies past a gap the tier narrows around the zones moves right or down by the room the tier took, up to what a fresh layout keeps there, so a load balancer between two zones keeps its lines 32 px or more off their borders.

fixed

Architecture: AWS edge-only edits make room for labels

Architecture: an edit to an AWS diagram that only adds edges gets the room check an edit that places nodes gets when one of the edges ends in the Region's column: a labeled line from a VPC, or a line that crosses a VPC's labeled line into the column. When a labeled run into the column is left too little room, the column moves right as one, with what lies right of the Region, and the edit's changed lists the moved nodes, where before the column stayed put and the new line could crowd the label. The check covers only edges that end in the Region's column.

fixed

Every link goes straight to its page

Every link on the site, in the docs, in Zindex's emails and in the files agents read opens its page directly, with no redirect on the way, and every page declares one canonical address, so search engines list each page once instead of as a page reached through a redirect. The signed-in pages, the password and verification pages and a saved diagram opened in the playground stay out of search results. The use cases, now the examples, and the element types reference, now under the docs, redirect to where they went, /dashboard and /settings open their first tab in one hop, a playground example that doesn't exist answers a real 404, and old links in emails already sent and on other sites keep working.

fixed

Example before and after pictures match the diff

On the four examples whose Revision diff shows before and after pictures, those pictures are revision 1 and the revision 2 the example's edit saves, so nodes the edit didn't touch stay where they were, as the diff says, and the main picture, the example's .svg, its thumbnail and its preview card show the same after picture. Until this release the after picture was the scene laid out from scratch, which moved nodes the edit kept, such as CloudWatch in the AWS example and every table and service in the ER and architecture diff examples. The main picture's caption says it is the saved revision, since the Playground draws the scene from scratch and its nodes can land elsewhere.

fixed

Mistyped verification codes keep you on the page

A wrong code on the email verification page says how many attempts are left and leaves you there to try again, where before one mistyped digit sent you to sign in again with "Your session expired". While a code is being checked, Resend code and signing out wait, and while a new code sends, Verify waits, so a resend can't replace the code being checked. Once the hourly limit on new codes is reached, Resend code reads "Limit reached - try again later" until the wait the limit sets has passed, and then works again.

fixed

Verification codes AutoFill into all six boxes

On the email verification page, a code filled into the first box in one go, as iPhone's AutoFill suggestion does, spreads across all six boxes and submits once, where before only its first digit was kept. After a paste or fill shorter than six digits, typing carries on at the first empty box.

fixed

Closing an account warns about lone teams' invoices

A team workspace only you are in is deleted with your account, and the Delete account dialog on the Account settings page, the email with the confirmation link and the workspaces docs say to download any invoices you need from it first, from Manage subscription on its Billing page, since its billing history can't be opened afterwards. The dialog and the email say this applies to a team workspace with billing history, and the email names only the team workspaces that have it. They already said the same for your personal workspace.

fixed

Provider buttons wait while the sign-in form sends

On the sign-in and sign-up pages, Continue with GitHub and Continue with Google can't be used while the email form is sending, as the form's own button can't, and work again if it fails. While a GitHub or Google sign-in opens, the form waits too, so one sign-in can't start on top of another, and if you stop the page from loading, the buttons and the form work again 12 seconds after the click.

fixed

Dialogs keep focus inside and return it

Every dialog on the site, among them the playground's Publish and Report a problem, the Account, Storage and Workspace settings, a diagram's Share, Report, Move and Delete, and a shared diagram's Report, keeps the keyboard inside it while it is open, and Escape and Cancel wait while its request runs. When it closes, focus goes back to the button that opened it, or to the next control when that button is gone, such as Revoke public link after a publish, and after an error to the button that failed. A click outside the playground's dialogs or a shared diagram's Report no longer closes them, and every dialog's title and buttons share one size.

fixed

Screen readers read success messages once

When a form succeeds and focus moves to its success message's heading, such as after sending a password reset link, a contact message or an invitation, a screen reader reads the message from that focus alone, where before it was also announced as a status update and could be read twice or cut short.

fixed

Team's plan details read the same everywhere

The pricing FAQ, the limits and workspaces docs and the OpenAPI spec's maximumAllowed say a Team workspace's limits are Pro's for each paid seat, at least 2 seats' worth, and the pricing FAQ says a viewer can hold 3 read-only API keys of their own, outside the pool. The pricing page's agent view says creating a private share link needs Pro or Team, where it said Pro alone. At a Team workspace's limit, the plan-limit message in the dashboard says each owner or member past the first two adds to its limits.

fixed

Sign-in and account pages share one layout

The pages for signing in, signing up, resetting a password, verifying your email, contacting Zindex, accepting an invitation, joining a workspace and connecting an app share one set of styles, with text aligned left and titles, lead lines and buttons at one size. Forgot password, reset password and contact replace the form with the site's success card once it sends, and Forgot password's card offers Send another link, which keeps your address.

fixed

The playground's Publish errors sit above its buttons

In the playground, an error from Publish shows above the dialog's buttons, as one from Report a problem does. A saved diagram's header uses the site's title and button sizes, so its title matches the playground's own, an error from Revoke public link shows in red rather than gray, and on a narrow phone the buttons wrap instead of pushing the page sideways.

fixed

Error messages no longer push buttons down

Errors on Billing's plan cards, the New team workspace page and the pages that confirm deleting an account or a workspace show in a red line kept free for them, as on the pricing page, so a one-line error no longer moves the buttons below it. Before, a refusal on Billing pushed its card's buttons down about 50 px, and Confirm deletion dropped 56 px.

fixed

Back to the Diagrams list keeps your place

Going back to the dashboard's Diagrams list after scrolling down it and opening a diagram returns to where you left the list, where before it could stop near the top because the list was still loading.

fixed

Wrapped page titles keep their lines together

A page title that wraps onto two lines, such as Confirm account deletion on a phone, keeps its lines close, where before a wide gap opened between them. Every page's title also takes less height, so the content under it starts slightly higher.

fixed

Diagram titles keep room in narrow windows

The Diagrams list's List view and the Recently deleted page switch to their phone layout at 640 px wide instead of 560 px, and below 768 px the Diagrams list hides Last changed, as phones already did. A diagram's title keeps 300 px or more at every width above a phone, where in a narrow window or a phone held sideways it could shrink to about 100 px.

fixed

Emails keep their font, size and color

Every email Zindex sends, among them verification codes, account-deletion confirmations, team invitations, workspace notices and inactivity warnings, shows its body text in the font, size and color it was designed with. In many email clients the text had lost all three, often showing in the client's default serif font.

breaking

Mermaid flowcharts import as flowchart diagrams

A Mermaid flowchart or graph creates a flowchart diagram, where it created an architecture one, drawn with the flowchart conventions in the direction its text gives, or top to bottom without one; to draw a service map as an architecture diagram, send an updateScene that sets diagramFamily to architecture after saving. Flowchart text edits a flowchart, an architecture diagram or a diagram without a family, each keeping its family, and an erDiagram edits an entity-relationship diagram, as before. Since Mermaid text names no family, a Mermaid read of any family but flowchart and entity-relationship carries the note FAMILY_NOT_EXPORTED: a diagram created from that text is a flowchart, while the saved diagram keeps its family.