Semantic forgetting

Semantic forgetting is the runtime's lifecycle layer for deciding when execution memory should stay active, become colder, move into archive, or be restored on demand.

What forgetting means here

Aionis treats forgetting as execution-memory maintenance. The runtime records whether a memory node should stay hot, move toward colder tiers, relocate payloads, or only be restored through a partial or differential rehydration path.

retain / demote / archive / reviewarchive relocationdifferential rehydrationsummary surfaces

Decision

Semantic forgetting

Lifecycle operators compute whether a node should be retained, demoted, archived, or pushed into review rather than treated as equally active forever.

slots.semantic_forgetting_v1

Relocation

Archive planning

Archive relocation captures whether payloads should stay local, move toward cold storage, and which payload scope is affected.

slots.archive_relocation_v1

Runtime

Rehydration control

When archived memory is needed again, Aionis can bring back only what is required through summary, partial, full, or differential rehydration.

/v1/memory/archive/*

Observation

Summary surfaces

Planning and execution introspection now expose forgetting, relocation, and rehydration summaries directly instead of hiding them only in raw node slots.

planning_summary.forgetting_summary

demotereviewarchiverehydrateThe important distinction is lifecycle control, not deletion.

Mental model

The forgetting path is useful because it explains why memory stays hot, why it becomes colder, and how the runtime brings it back when the task needs that context.

What is active today

Today the runtime already supports:

1. lifecycle operators that write semantic_forgetting_v1
2. archive relocation plans that write archive_relocation_v1
3. differential payload rehydration for anchored workflow memory
4. planning and execution summary surfaces that expose these signals directly

That means forgetting is visible through public runtime behavior.

The main public entry points

SDK method	Route	What it proves
memory.archive.rehydrate(...)	POST /v1/memory/archive/rehydrate	Archived nodes can be brought back into the active working tier
memory.nodes.activate(...)	POST /v1/memory/nodes/activate	Reused nodes record whether they actually helped
memory.anchors.rehydratePayload(...)	POST /v1/memory/anchor/payload/rehydrate	Payload restoration can stay selective instead of forcing full restore
memory.planningContext(...)	POST /v1/memory/planning/context	Planning can show when colder memory should stay out of the default working set
memory.executionIntrospect(...)	POST /v1/memory/execution/introspect	Execution introspection can show archive and rehydration state directly

What the new surfaces expose

The forgetting summaries now include:

• semantic action counts
• lifecycle-state counts
• archive relocation state, target, and payload scope counts
• rehydration mode counts
• differential rehydration candidate counts
• a runtime-level recommended action

Those fields exist to answer questions such as:

• why was this memory demoted instead of kept hot?
• why is this workflow archived instead of reused immediately?
• why did the runtime recommend widening recall or rehydrating a colder payload?

What to look for when you evaluate it

The easiest signals to inspect are:

1. whether planning explains why colder memory stayed out of the default working set
2. whether execution introspection shows archive and relocation state directly
3. whether rehydration restores only the level of detail the task needs

Practical reading order

If you want to understand this surface quickly:

1. read the Memory page
2. run the semantic forgetting proof demo
3. inspect planning_summary.forgetting_summary
4. inspect execution_summary.forgetting_summary
5. then read Proof by Evidence

Runnable proof

Run:

npm run example:sdk:semantic-forgetting

What this demo proves:

1. a cold workflow anchor can be archived without being deleted
2. planning can explain why hotter memory is still preferred first
3. execution introspection can expose archive and relocation state directly
4. differential payload rehydration can restore only the archived detail that is needed
5. archive rehydration can move the workflow back into the active tier without erasing its colder-memory recommendation