PAIOS CE v0.1.0-rc.2

PAIOS CE v0.1.0-rc.2

PAIOS CE is a local-first execution-control reference runtime for AI
agent builders.

This release candidate focuses on making the execution decision explicit,
auditable, and visible before a scoped action proceeds.

Highlights

• explicit execution-control and permission-gate flow
• real blocked-action demo with zero execution attempted
• real allowed-action demo for scoped low-risk execution
• deterministic architecture and security-boundary visuals
• generated configuration schemas aligned with the current runtime
• contribution, security, support, Issue Form and PR infrastructure
• English and Simplified Chinese project documentation

Verified behavior

Blocked unknown action:

• decision: blocked
• error_code: E_UNKNOWN_ACTION
• execution_attempted: false
• exit code 5

Allowed demo:

• decision: allow
• risk_level: LOW
• successful result
• exit code 0

Validation

• model contract: 9/9 passing
• regression suite: 298 passed, 1 filesystem-specific skip
• line coverage: 87.83%
• branch coverage: 76.62%
• fresh wheel/sdist build passes
• fresh install and CLI smoke checks pass
• no confirmed secrets, PII, or private machine paths in the public release candidate

Current scope

PAIOS CE is a reference runtime and release candidate.

Current public scope does not claim production support for:

• live WeChat or Feishu channels
• production cloud-model providers
• browser automation
• arbitrary shell / administrator execution
• remote desktop control
• production OpenClaw or Hermes integration

See the repository README, security model and roadmap for current
capabilities and planned work.

Documentation

• README
• Security policy
• Security model
• Contributing guide
• Roadmap

Package version

Python package version:

0.1.0rc2

This GitHub release is a pre-release.

PAIOS CE v0.1.0-rc.1

PAIOS CE v0.1.0-rc.1

Project

PAIOS CE is a clean-room public core scaffold for a privacy-preserving personal AI orchestrator. It implements the public skeleton, interface contracts, and a minimal local demo closed loop from scratch, in full compliance with the frozen architecture.

Core capabilities

• Clean-room architecture — newly written from public contracts; no legacy source, no production data, no owner PII.
• Secret management — SecretRef reference-only indirection; secret values are never stored in config, logs, error messages, audit events, or envelopes.
• Permission boundary — deny-by-default permission engine with capability allowlists, filesystem/network scope, HIGH-risk confirmation, and CRITICAL/ADMIN/shell hard blocks.
• Offline validation — fully local Mock Provider (deterministic, no network, no API key); offline test and audit pipelines.
• Security-first agent infrastructure — redacted audit logging, loopback-only defaults, dm-only with group execution off, owner allowlist required.

Validation

• 203 tests passing — 0 failures / 0 errors / 0 skipped
• Python 3.11 and 3.12 validated on Windows and Ubuntu (4 matrix jobs)
• Secret pattern scan passed
• pip-audit passed — 0 known vulnerabilities
• Dependency audit passed (installed dependency inventory frozen as a CI artifact)

Repository governance

• Protected main branch — 16 required status checks (CI / Package / Security), strict up-to-date, enforce-admins, force push disabled, deletion disabled.
• Restricted GitHub Actions — GitHub-owned actions only; GITHUB_TOKEN default read; pull-request create/approve disabled; zero repository secrets.
• Reproducible build validation — wheel and sdist were built, checked, and clean-install tested in GitHub Actions; this pre-release currently exposes GitHub's automatic source ZIP and tar.gz archives.

Current positioning

This is a release candidate.

It is not a final stable release. Mock Provider and CLI are EXPERIMENTAL. No live Provider, live Channel, Speech, Hermes, or OpenClaw integration is shipped. Publication of PyPI/Zenodo is not included in this release.