Epochs

An epoch is a bounded, sealable segment of work for lifecycle management. You open an epoch, do work inside it, then seal it — capturing a Merkle root of its contents so the sealed segment is immutable and tamper-evident. Epochs are how you turn a stretch of agent activity into an auditable, frozen record (a release, a review window, a billing period) without stopping the live graph.

Available over MCP and the HTTP API.

Lifecycle

// agentstategraph_create_epoch
{ "id": "2026-q3-release", "description": "Q3 release window" }

// agentstategraph_enter_epoch    → subsequent work is attributed to this epoch
{ "id": "2026-q3-release" }

// agentstategraph_exit_epoch     → stop attributing to it (still open/unsealed)
{ }

// agentstategraph_seal_epoch     → make it immutable
{ "id": "2026-q3-release", "summary": "Shipped 0.9.21; 42 commits, 3 policy changes" }

Sealing records the seal time and a Merkle root of every object in the epoch, computed from the then-current main chain. After sealing, the segment cannot be modified — any later tampering is detectable by recomputing the root.

Inspecting & archiving

// agentstategraph_list_epochs    → all epochs with status (open / sealed)
{ }

• agentstategraph_export_epoch — export a (typically sealed) epoch as a portable record for audit or handoff.
• agentstategraph_archive_epoch — move a sealed epoch out of the active set once it’s no longer day-to-day relevant.

When to use epochs vs. branches

• Branches isolate concurrent lines of work you intend to merge back.
• Epochs freeze a completed slice of history into an immutable, verifiable record. Reach for an epoch when you need a defensible “this is exactly what happened during window X” — releases, audits, compliance windows.

See also

• Namespaces — the per-tenant isolation unit epochs live within
• Policy — governance rules evaluated as work enters the graph