Canvas Memory

Canvas memory stores structured JSON data on a canvas, scoped by namespace. It persists across runs so workflows can keep state, pass data between branches, or look up records by matching fields.

How it works

• Canvas-scoped — Data belongs to one canvas.
• Namespaces — You choose a namespace string per use case (for example deployments or incidents).
• Rows — Each write can add or update rows; each row holds JSON values (object).
• Matching — Read, update, delete, and upsert operations can match rows by field values (JSON containment).

Persisting data across runs

Runs on the same canvas are independent: payloads from one run do not automatically appear in the next. Canvas memory is where you store facts that should survive until a later run needs them.

Pattern	How
Staged rollout	Upsert the current stage; next run Read (latest) to decide whether to advance.
Incident handoff	Run 1 Add IDs; Run 2 Read by key to continue the same incident.
Deduplication	Upsert on a stable key; branch on found / notFound.
Counters	Update to merge new values (e.g. attempt count); later runs read the totals.

Components

SuperPlane provides five components for canvas memory:

Component	Purpose
Add Memory	Append a new row in a namespace.
Read Memory	Find rows by namespace and match criteria; emits on found or notFound.
Update Memory	Update matching rows by merging new fields; emits on found or notFound.
Delete Memory	Delete matching rows; emits on deleted or notFound.
Upsert Memory	Update the first match, or insert if none match.

Read Memory supports:

• Result mode — all (every match) or latest (most recent).
• Emit mode — allAtOnce (one event with all matches) or oneByOne (one event per row).

See the Components reference for field-level configuration.

API

Canvas memory can also be listed or cleared via the REST API (see Public API Reference). Managing memory records requires canvas permissions that allow reading or updating the canvas as appropriate.