Per Ardua

Open Source

Tools for AI research, organizational analysis, and privacy-preserving infrastructure

github.com/jmcentire

1

Reference Impl

6

AI & Agents

5

Stack Governance

3

Org & Management

3

Infrastructure & Privacy

Reference Implementation

Exemplar

The governance stack in practice — not a demo, but the actual product

Governed code review service — the reference implementation for the entire governance stack. Every component specified by Constrain, decomposed by Pact, routed by Baton, trust-scored by Arbiter, and sealed by Tessera. Multi-stage analysis (security, correctness, style, architecture) with full audit trails. 1,979 tests across 43 test files.

Python GitHub

AI & Agents

Agent authorization, model distillation, and multi-agent coordination

Signet

Personal sovereign agent stack. Gives AI agents a cryptographic vault for managing user credentials, generating zero-knowledge proofs, and enforcing privacy policies. 62K lines, 1049 tests, MCP-native.

Rust

Signet-Eval

Deterministic policy enforcement for AI agent tool calls. Aviation-inspired red lines with plan B alternatives, preflight self-planning with timed lockout, spending limits, credential gating, encrypted vault. Three enforcement tiers: pause, disable, normal. 22 MCP tools, 6 locked self-protection rules, 97 tests including adversarial.

Rust

Agent-Safe

Embed authorization in the token. 150 lines. Zero deps. Microseconds. Defines SPL (Safe Policy Lisp), a tiny S-expression policy language that travels inside signed capability tokens. 6 SDKs.

Java Rust Go Python TypeScript C#

Pact

Contracts before code. Tests as law. Agents that can't cheat. Multi-agent software engineering framework with phase-aware health monitoring, Goodhart adversarial tests, and processing register propagation. 100% on ICPC World Finals. 1814 tests, on PyPI.

Python

Apprentice

Adaptive model distillation with story-level learning. Routes requests between frontier API and local model, progressively shifting traffic as quality proves out. Journey evaluator scores multi-step flows. Per-journey phase tracking. 2791 tests.

Python

Constrain

Find the boundary between specification and intent. CLI tool that interviews engineers about their problem and produces an induced-understanding prompt and structured constraint set. Randomized adversarial postures ensure descriptions are robust.

Python

Stack Governance

Trust enforcement, schema governance, production attribution, and stack adoption for Pact-built systems

Baton

Cloud-agnostic circuit orchestration. Topology-first service mesh with OTLP-native observability, multi-endpoint span fan-out, reverse-proxy adapters, taint analysis, canary auto-promotion, and research-backed self-healing. 983 tests.

Python

Arbiter

Access auditing, consistency analysis, blast-radius classification, and trust enforcement. Append-only trust ledger with SHA-256 checkpoints. Five-factor trust formula. Three-step conflict resolution: authority, trust arbitration, human escalation.

Python

Sentinel

Production attribution and contract tightening. Watches logs for errors, attributes them to components via PACT keys, spawns LLM fixer agents, pushes tightened contracts back to Pact, and emits incident stories to Chronicler. Multi-window budget caps.

Python

Ledger

Schema registry and data obligation manager. Registers schemas across 12 backend types, validates field-level annotations, gates migrations (BLOCKED/HUMAN_GATE/AUTO_PROCEED), and exports rules to Pact, Arbiter, Baton, and Sentinel.

Python

Cartographer

Stack adoption and compatibility tool. Scans existing codebases via AST analysis, produces draft artifacts for all stack tools with confidence levels, and runs 34 compatibility checks across 5 tools. Read-only invariant.

Python

Org & Management

Stigmergic coordination, knowledge graphs, and management tools

Chronicler

Event collection and story assembly. Correlates raw events into stories at configurable granularities (request, service, journey) and emits them to Stigmergy for pattern discovery and Apprentice for learning. Closes the product feedback loop. 1272 tests.

Python

Stigmergy

Organizational awareness through stigmergic signal processing. Ingests work artifacts from GitHub, Linear, and Slack, routes them through an ART mesh with sequence-aware familiarity scoring, and surfaces structural patterns including normalized deviance. 1545 tests.

Python

Ascend

AI-powered engineering management CLI. Track your team, analyze performance, prepare for 1:1s, and detect risks from the terminal. Roster management, meeting transcript parsing, risk detection (flight risk, burnout, bus factor), coaching, OKR/PIP tracking, and interactive TUI.

Python

Kindex

The memory layer Claude Code doesn't have. Persistent knowledge graph for AI-assisted workflows. Indexes conversations, projects, and intellectual work so Claude Code never starts a session blind. 591 tests, on PyPI.

Python

Infrastructure & Privacy

Document formats, encrypted messaging, and privacy architecture

Tessera

Self-Validating Executable Document Format. A document that proves itself. Carries its data, schema, mutation history, and cryptographic signatures in a single portable artifact. Ed25519/SHA-256 hash chain, embedded Rhai scripting, 130+ tests.

Rust

HermesP2P

Decentralized, ephemeral peer-to-peer messaging. No servers, no persistence, no compromise. Transient communication network with cryptographically secure messaging overlay, PGP-style encryption, public/private channels.

JavaScript

BlindDB

Interactive demo of client-side encryption architecture from The Ephemeral Internet. Shows two architectures side by side: today's 12-exchange credential flow vs. tomorrow's 2-exchange flow where the server stores only encrypted blobs at opaque addresses.

HTML