What makes an AI diagram persistent?

It has a stored diagram id and stable editor URL in an account workspace. People and authorized programmatic clients return to that identity, so later edits, comments, access, and renders operate on the same artifact.

Is the diagram id a secret?

The authenticated editor and API still enforce access, so the id is not a bearer credential. Treat internal diagram references as private workflow context anyway, and never store PATs or other credentials beside them in a public repository.

How should I share a persistent diagram?

Invite a collaborator when they should sign in, comment, or edit according to their role. Enable a public link only for anonymous read-only viewing, optionally with a password, and disable it when that access is no longer needed.

Does reopening this example prove retention?

No. It proves that the same artifact can be reopened in a later session. Retention is a user-behavior outcome and must be measured from mature, privacy-reviewed usage rather than inferred from product capability.

Persistent AI diagrams · one artifact lifecycle

Keep the diagram after the chat ends

Treat the agent’s first draft as the start of an artifact, not the end of a response. Keep one diagram id and editor URL across browser corrections, teammate comments, targeted agent updates, and a deliberate later-session reopen.

Diagram id

Stable identity

Agent

Bounded draft

Human

Browser correction

Team

Comments + access

Return

Later-session reopen

Continuity comes from one owned diagram identity. The agent returns an id and editor URL; the browser edits that same canvas; collaborators attach access and comments to it; later tools and people return to the same id instead of reconstructing context from an exported image.

Product boundary: Reopening a stored diagram is a product capability, not a retention result. This walkthrough demonstrates one later-session reopen but does not prove that users return, that a team retained the workflow, or that persistence improves outcomes. Those claims require observed mature usage after an approved launch.

Reproduce it

Copy setup

Create one persistent diagram

Connect @excaliwow/mcp with read + write access and ask the agent to build the bounded architecture below. Keep the id and editor URL returned by the create tool; do not rely on the chat transcript as the only pointer.

MCP client config

{
  "mcpServers": {
    "excaliwow": {
      "command": "npx",
      "args": ["-y", "@excaliwow/mcp"],
      "env": {
        "EXCALIWOW_TOKEN": "excw_pat_…"
      }
    }
  }
}

Save the identity outside the chat

Copy the returned diagram id into an owner-only local file and print the editor URL. The id is not a bearer token, but storing project references outside a public repository avoids leaking internal artifact names and workflow context.

Terminal

export DIAGRAM_ID=<id-returned-by-create-tool>
ID_DIR="$HOME/.config/excaliwow/diagrams"
ID_FILE="$ID_DIR/saas-architecture.id"

install -d -m 700 "$ID_DIR"
printf '%s\n' "$DIAGRAM_ID" > "$ID_FILE"
chmod 600 "$ID_FILE"

printf 'Editor: https://excaliwow.com/c/%s\n' "$DIAGRAM_ID"

Reopen it in a later session

Authenticate the CLI, read the stored id, verify the current title and updated time, and print the same editor URL. In the browser, invite an editor or leave a pinned comment; use a public link only when read-only access is intended.

Later terminal session

npm install -g @excaliwow/cli
excaliwow auth login

ID_FILE="$HOME/.config/excaliwow/diagrams/saas-architecture.id"
test -s "$ID_FILE"
DIAGRAM_ID=$(tr -d '\n' < "$ID_FILE")

excaliwow diagrams get "$DIAGRAM_ID" --json \
  | jq '{id, title, updatedAt}'

printf 'Reopen the same canvas: https://excaliwow.com/c/%s\n' "$DIAGRAM_ID"

Human review

Change what the agent misunderstood

Correct a misplaced system boundary before polishing the layout.
Clarify one ambiguous data-flow label so reviewers do not infer unsupported behavior.

Same-canvas follow-up

Send the correction back

Update the same diagram in place with the reviewed relationship, preserving every human edit and existing element id.

Review the critical path, not the pixels

The create result, saved reference, CLI read, and editor URL all use the same diagram id.
Browser corrections and the later agent update land on the same live canvas rather than a replacement export.
A teammate invite, pinned comment, or read-only share remains attached to the diagram rather than the chat response.
The later-session command reopens the same id and inspects its current updated time.
The walkthrough is reported as capability proof, never as observed return or retention evidence.

Common questions

What makes an AI diagram persistent?: It has a stored diagram id and stable editor URL in an account workspace. People and authorized programmatic clients return to that identity, so later edits, comments, access, and renders operate on the same artifact.
Is the diagram id a secret?: The authenticated editor and API still enforce access, so the id is not a bearer credential. Treat internal diagram references as private workflow context anyway, and never store PATs or other credentials beside them in a public repository.
How should I share a persistent diagram?: Invite a collaborator when they should sign in, comment, or edit according to their role. Enable a public link only for anonymous read-only viewing, optionally with a password, and disable it when that access is no longer needed.
Does reopening this example prove retention?: No. It proves that the same artifact can be reopened in a later session. Retention is a user-behavior outcome and must be measured from mature, privacy-reviewed usage rather than inferred from product capability.

Keep one diagram identity through the review

Start free—no credit card. First 10,000 verified accounts keep the core workspace free; 3 diagrams to start and earn more capacity.

Start free