Skip to content

simulation

A simulation — an isolated runtime where an agent runs. The “VM” to a memex’s “disk image.” Loads a memex, applies governor rules, and runs an agent in a controlled environment.

Simulations provide isolation: writes go to a fork overlay, not the original memex. The user reviews changes and merges selectively — like a pull request for your knowledge graph.

Examples: research-agent (forked joe.db + astronomy.db read-only), legal-review (forked joe.db + law.db read-only)

MetadataValue
Pluralsimulations
Subtitle fieldstatus
FieldType
statusstring
profilestring
tasktext
graphModestring
startedAtdatetime
endedAtdatetime
actionCountinteger
writeCountinteger
RelationTarget
primaryMemexmemex
mountedMemexmemex[]
agentagent
tetherhardware
forkedFromsimulation
startedByperson

External standards this shape draws from or aligns with. See Shape design principles for how prior art informs shape design.

  • OpenTelemetry Traces (root span + attributes) — Span-shaped observation of an agent run. Our startedAt/endedAt/ actionCount/writeCount ≈ span attributes; status ≈ span status.
  • QEMU / VM snapshots — “Disk image vs. VM” metaphor is direct. Our primaryMemex ≈ writable disk; mountedMemex[] ≈ read-only overlays; forkedFrom ≈ snapshot-based fork.
  • Kubernetes Pod + Volume mounts — Our tether (hardware kill-switch) ≈ Pod security context; mountedMemex[] ≈ ConfigMap/PVC read-only mounts.