GitHub Download

v1.6.0 Apache 2.0 · Local · No telemetry

APXV Lab — design, bind, and prove agents locally

APXV™ is a local Lab and runtime. Create Bound Agents (Agent + Packs + Proof), equip tools and skills, wire Pipelines, compose Swarms, and verify fail-closed claims on infrastructure you control — air-gapped, no cloud trust boundary.

Download desktop Lab guide
  • Desktop installers (primary)
  • Bound Agent · Pipeline · Swarm
  • Tools · Skills · local MCP
  • API on 127.0.0.1:8741

About APXV 1.6.0

APXV (Attested Proof Execution Verified) is an open-source, air-gapped Lab and runtime for governed agents. Version 1.6.0 makes APXV Lab the primary product surface.

What you get in 1.6.0

  • Lab IDE — Agents, Packs, Proofs, Pipelines, Swarms, Runs, System
  • Bound Agents — loadout with packs + proof profile; Test → Bind
  • Official packs v0.1 — Redaction, Document Processing, AI Governance
  • Desktop installers — Windows MSI/NSIS, Linux deb/AppImage
  • Local API — 127.0.0.1:8741, no cloud account
  • Optional ZK — governance + entity Groth16 when you attest

Who it is for

  • Operators who need governed agents on their own hardware
  • Teams that must keep data and keys air-gapped
  • Builders equipping vertical packs without multi-tenant SaaS

Release notes · v1.6.0 · Changelog

Built for local control

Not a chatbot SaaS. A Lab for privacy-sensitive agent systems — Bound Agents, Pipelines, and Swarms — with policy, audit, tools, and cryptographic receipts on your machine.

  • Local by default

    Runtime and data under your managed/ tree. No cloud trust boundary and no telemetry.

  • Agent + Pack + Proof

    Agents wear packs (verticals) and a proof profile. Test under the loadout, then Bind to the runtime.

  • Honest claims

    Catalog proof tiers (structural · attestation · zk · vertical) fail closed. Dual-track Groth16 when you attest with real provers. Not “LLM correctness.”

  • Tools · Skills · MCP

    Sandboxed built-in tools, skill instruction bundles, and a local stdio MCP host (including a built-in smoke server). Bounded — not unrestricted shell access.

How it works

Install once, connect locally, then build systems you can re-run and verify.

  1. 01

    Install & connect

    Prefer the desktop installer. Paste your operator key, open Lab. Optional bootstrap for operator-owned proving keys.

  2. 02

    Bind

    Create an Agent, equip Packs and a Proof Profile, optionally tools and skills, Test under that loadout, then Bind.

  3. 03

    Compose

    Place agents and pack steps on a Pipeline; link pipelines on a Swarm. Run.

  4. 04

    Prove

    Read fail-closed claims on the run. Attest when you need dual-track Groth16; verify artifact hashes in Trust / Runs.

What ships in Lab

Primary product surface. Direction and honesty live in the public roadmap.

  1. Shipped

    Atoms

    Agents, packs (verticals), and proof profiles — install, equip, attest, verify.

  2. Shipped

    Bound Agents

    Loadout of packs + proof (+ tools/skills); Test → Bind; maturity draft / tested / bound.

  3. Shipped

    Pipelines & Swarms

    Compose agent/pack graphs; link pipelines as one system; run strip with fail-closed claims.

  4. Shipped

    Tools · Skills · MCP

    Sandboxed built-ins, skill bundles, local stdio MCP host with a Lab smoke server.

  5. Later

    Ecosystem

    Optional share/import and community registry — not required for local use.

Not building: cloud multi-tenant SaaS, compliance certification product claims, unrestricted remote code execution, or free-form ungoverned agent graphs.

Platform surface

Open-source runtime and Lab UI (v1.6 line).

  • Desktop + Lab UI

    Windows and Linux installers are the primary operator path. Lab: agents, pipelines, swarms, runs, trust, tools/MCP.

  • Local API v2

    /api/v2/* for agents, packs, proofs, pipelines, swarms, jobs, lab tools/MCP, governance, health.

  • Bound Agents

    Packs and proof on the loadout; Test under loadout; Bind to signed capabilities.

  • Chained audit

    Hash-linked logs for system and agents. Doctor and integrity repair.

  • Dual-track ZK

    Governance + entity Groth16 when you attest. Catalog claim tiers stay fail-closed.

  • Bring your own LLM

    Optional local backends (Ollama). No bundled model weights. Chat is optional for composition.

Includes

  • APXV Lab (primary surface)
  • Bound Agents, Pipelines, Swarms
  • Built-in tools, skills, local MCP host
  • Official packs v0.1: Redaction, Document Processing, AI Governance
  • Doctor, integrity checks, install scripts
  • Optional sovereign bootstrap for operator-owned keys

Does not include

  • HIPAA / SOC2 / GDPR certification
  • Cloud SaaS or multi-tenant hosting
  • Proof that an LLM answer was “correct”
  • Arbitrary browser-authored ZK circuits
  • Unrestricted network/shell tools
  • Enterprise DLP product claims

Threat model

Packs are verticals

A pack is a vertical an agent wears — policy, knowledge, and sometimes a runnable path. Equip what you need; share one APXV instance.

  • Redaction Pack

    Official · v0.1 — Governed text redaction, orchestration, and dual-track attestation.

    View pack

  • Document Processing Pack

    Official · v0.1 — Batch .txt / .json ingest with compliance policy for document folders.

    View pack

  • AI Governance Pack

    Official · v0.1 — Redaction plus LLMReasoner review under confidence and cost constraints.

    View pack

These three are the Official packs at v0.1. Equip them in Lab on Bound Agents, or follow Build your first pack for custom packs.

Download

Desktop installers are how APXV is meant to be used. Get the latest build from GitHub Releases, then operator key → Open Lab. Optional sovereign bootstrap for operator-owned keys.

Windows

MSI or NSIS setup · x64

APXV_*_x64-setup.exe

Linux

amd64 .deb or AppImage

APXV_*_amd64.deb

macOS

DMG planned for a follow-up release

—
GitHub Releases Install guide Downloads hub

Documentation

Start with the Lab guide. Full detail lives in the repository on GitHub.

README · Changelog · Migration v1.6 · Issues

Contact

Bugs and how-to: open a GitHub issue and include python -m scripts.apxv_doctor output. Security: see SECURITY.md.

GitHub Issues APXVdev@protonmail.com @APXVdev