VMem

Forge Pool VMem provides shared execution working memory for distributed workloads executed by the Forge Pool platform.

VMem stores reusable computational assets generated during execution while preserving the separation between immutable execution contracts and transient execution state.

Rather than acting as persistent storage, VMem serves as the shared working memory layer of the execution platform.

---

Purpose

Distributed execution frequently produces intermediate computational assets that remain valuable beyond a single execution step.

Rather than recomputing these assets repeatedly, VMem allows them to be reused throughout larger execution workflows.

Representative uses include:

• reusable computational baselines
• intermediate execution outputs
• numerical state
• parameter grids
• execution checkpoints
• workflow handoff payloads
• cached execution surfaces

VMem supports computation.

It does not define computation.

---

Execution Working Memory

Execution contracts remain immutable.

Execution working memory does not.

VMem stores transient computational state that supports active execution without changing execution semantics.

Execution truth remains defined by the execution contract.

VMem supports execution efficiency.

---

Position in the Execution Platform

VMem participates throughout distributed execution.

Execution Contract
        ↓
Distributed Execution
        ↓
Execution Working Memory (VMem)
        ↓
Reduction
        ↓
Execution Evidence

Working memory supports execution while remaining independent of execution semantics.

---

Runtime Characteristics

VMem is characterized by:

• shared execution working memory
• reusable computational assets
• execution-scoped storage
• transient lifecycle
• version-aware payloads
• replay-aware references
• workflow-oriented reuse

---

Conceptual Operations

VMem supports four conceptual operations.

• Store
• Load
• Release
• Delete

Representative API endpoints include:

POST   /v1/vmem/store
POST   /v1/vmem/load
DELETE /v1/vmem/delete

The execution model remains independent of the transport interface.

---

Execution-Aware Lifetime

VMem is intended for active execution.

Stored payloads commonly exist only for the duration required to support ongoing computational workflows.

Long-term execution evidence belongs elsewhere.

VMem therefore prioritizes execution reuse rather than permanent persistence.

---

Runtime Guarantees

VMem inherits the execution philosophy of the Forge Pool platform.

Execution working memory remains:

• execution-scoped
• lifecycle-managed
• version-aware
• replay-aware
• deterministic in addressing
• suitable for distributed execution workflows

---

Relationship to KV

KV stores lightweight execution state.

VMem stores reusable computational payloads.

Together they provide complementary execution surfaces.

KV answers:

"What is the execution state?"

VMem answers:

"What computational assets should active execution reuse?"

---

Relationship to Replay

Replay references may identify payloads previously stored in VMem.

Replay itself remains defined by execution evidence rather than working memory.

VMem supports replay.

It does not define replay.

---

Relationship to Verification

Verification may inspect computational assets generated during execution.

VMem therefore supports verification workflows without determining verification outcomes.

Verification validates execution.

VMem provides reusable computational context.

---

Relationship to Execution Artifacts

Execution artifacts preserve long-term execution evidence.

VMem preserves transient computational assets.

Artifacts support trust.

VMem supports execution efficiency.

---

What VMem Is Not

VMem is intentionally not:

• a relational database
• an object storage platform
• an artifact repository
• a distributed filesystem
• a long-term persistence layer

Its purpose is shared execution working memory.

---

Continue in Forge Studio

Observe execution working memory by:

• inspecting reusable execution payloads
• reviewing intermediate computational assets
• exploring workflow composition
• comparing replay references
• examining execution reuse across workloads

Trust should be established through independent verification rather than documentation alone.

---

Final Note

VMem is not simply shared memory.

It is the execution working memory surface of the Forge Pool platform.

By separating reusable computational assets from immutable execution contracts and long-term execution evidence, Forge Pool enables efficient distributed execution while preserving deterministic behavior, replay compatibility, verification support, and architectural clarity.