Skip to content

Repository files navigation

Persisting

Agent execution, orchestration, and durable history.

Persisting is an Agent infrastructure toolkit built from three peer products:

  • pVisor runs one Agent in a reviewable workspace;
  • pPilot plans, executes, and recovers many independent Runs;
  • pChronicle discovers, queries, exchanges, and serves trajectory Datasets.

Gateway, Control, OverlayFS, and OverlayNet are runtime drivers assembled by pVisor. Queue, Search, and Tensor Memory are separate data capabilities; they are not dependencies of the Agent execution path.

pPilot ── RunSpec ──► pVisor ── captured events ──► pChronicle
  │                    │                              ▲
  └──── results ───────┴──────────────────────────────┘
                       ├─ Control
                       ├─ OverlayFS
                       └─ OverlayNet → Gateway

Install

The Python wheel installs the Python package and a matched set of three host commands:

pip install persisting

pvisor --version
ppilot --version
pchronicle --version

The rolling nightly build is available without a Rust toolchain:

curl -fsSL https://raw.githubusercontent.com/DeepLink-org/Persisting/main/scripts/install-nightly.sh | bash

Source developers can install the same command set with just install-cli. See the installation guide for platform requirements and executor setup.

Start with the task

Run and review one Agent

pvisor run --safe codex
pvisor review last
pvisor apply last       # or: pvisor drop last

--safe stages workspace changes. The exact filesystem and network boundary is platform-dependent and is recorded in the Run Bundle; consult the pVisor guide before treating it as a security boundary.

Orchestrate many Runs

Create a Python plan with plan() and execute(item), then run it with bounded parallelism and a durable result journal:

ppilot run plan.py --workers 4 --per-worker 2 --sink ./results

For independent pVisor workspaces, use ppilot produce. Dataset queries and analysis belong to pChronicle.

Browse and analyze trajectories

# Optional: configure one local Dataset root as the default Warehouse.
pchronicle default ./trajectory-data

pchronicle ls
pchronicle status
pchronicle analysis overview
pchronicle query "SELECT agent_id, COUNT(*) AS runs FROM dataset.runs GROUP BY agent_id"

An explicit local path or S3 URI can be supplied instead of configuring a default Warehouse. pchronicle import and export exchange ATIF, ACTF, OpenAI Messages, and Storyline JSON; pchronicle serve starts the loopback-only, read-only Warehouse UI and API.

Command ownership

Command Primary responsibility
pvisor One Run, environments, review, checkpoints, apply/drop
ppilot Batch planning, bounded execution, recovery, and Run production
pchronicle Dataset catalog, SQL, built-in analysis, find, import/export, read-only serving

Current maturity

Capability Status
pVisor host execution, review, checkpoints, and transactional workspace Implemented
pPilot batch orchestration and durable recovery Implemented
pChronicle local/S3 catalog, bounded SQL, analysis, find, import/export Implemented
pChronicle loopback-only read API and embedded Web UI Implemented
Gateway capture and cooperative proxy policy Implemented
Container/libkrun executors and transparent network boundaries Platform-dependent; see the pVisor and OverlayNet docs
Queue and document Search Separate stable capabilities
Tensor Memory / TTAS Experimental

The CLI --help, component READMEs, tests, and user guides describe supported behavior. Files under docs/src/design/ may also contain explicitly labelled target architecture; RFCs preserve decisions and are not command references.

Documentation and examples

From a source checkout:

just examples
just docs-build
just docs-links

License

Apache License 2.0. See NOTICE for third-party attributions and separately licensed bundled components.

About

Persistent Storage for Parameters, KV Cache, and Trajectories

Resources

Stars

1 star

Watchers

1 watching

Forks

Releases

Packages

Contributors

Languages