Context & Memory

The Agentic Framework distinguishes per-execution context from persistent memory.

Execution context (per-execution)

Owned by for the duration of a single execution. Holds:

Field	Notes
`inputValues`	Frozen — provided at execution start
`stateValues`	Mutable — initialized from `state_variables` + defaults
`outputValues`	Set on `end` node
`nodeOutputs`	Map of `nodeId → output`
`executionTrace[]`	Append-only audit log of node starts / completes / errors
`currentNodeId`	Current node
`status`	`pending` / `running` / `completed` / `failed` / `paused`
`startedAt`, `completedAt`	ISO timestamps	Resolution is via `resolveVariable(expression)`: e.g. `{{nodes.intake_1.output.transcript}}`.

Memory (persistent)

Persistent agent memory is stored under the tenant scope. Memory survives across executions and contexts and is read by agents that bundle the Memory Expert (see /agentic/experts/memory).

Scope	Lifetime
`thread`	Until context ends
`patient`	Until patient deletion
`tenant`	Indefinite (subject to HIPAA 6-year retention)

​Execution context (per-execution)

​Memory (persistent)

Execution context (per-execution)

Memory (persistent)