pVisor

pVisor is the AgentVisor implementation in Persisting. It virtualizes Agent execution: shared host, container, VM, and future fleet resources are presented to each Run as an isolated Agent virtual execution environment.

What pVisor owns

• a stable Run identity independent of the physical process or provider;
• creation, admission, cancellation, recovery, checkpoint, and terminal state;
• workspace, network, tool, model, credential, and compute capabilities;
• containment and review of effects where the medium supports it;
• evidence describing the controls that were actually installed.

pVisor does not define the Agent reasoning loop. It runs existing Agent CLIs, scripts, and frameworks inside a governed execution boundary.

Begin with one Run

pvisor run --safe codex
pvisor review last
pvisor apply last --path src

The Agent can edit freely inside its stage. The user decides which filesystem effects enter the base project, and can apply independent batches more than once.

The standalone pVisor product loop is:

RunSpec -> admission -> Attempt
  -> terminal RunResult + private Run Bundle + staged Effects
  -> later review/apply/drop

Attempt finalization writes the terminal RunResult and private, versioned Run Bundle while leaving filesystem Effects staged. Later review, apply, or drop operations read the Bundle and operate on the stage. pChronicle is not a runtime prerequisite for this loop.

Read pVisor by purpose

Goal	Section
Complete the first local Run	Get Started
Understand the category and object model	Concepts
Choose an executor or govern effects	Guides
Inspect isolation and runtime mechanisms	Design
Look up exact command syntax	Reference

To query configured Gateway trajectory events and pVisor lifecycle records as a durable Dataset, continue to pChronicle. The current handoff does not publish the full Run Bundle or its Artifact, lineage, Effect, and broader Evidence inventory.