Making good agents better — safe, auditable, and verifiable — without making them slow.

---

Lightweight, deterministic, policy-bound, auditable, and verifiable execution runtime for AI agents operating in regulated environments.

Reference domain: Healthcare

Quick Start

git clone https://github.com/Chesterguan/veritas.git
cd veritas
cargo test --workspace       # 58 tests, all passing
cargo run -p demo -- run-all # run all 5 healthcare scenarios

Or launch the interactive TUI:

cargo run -p veritas-tui

The TUI lets you select scenarios, toggle patient consent and agent capabilities, and watch VERITAS enforce policy in real time.

Prerequisites: Rust 1.74+ (install)

Why VERITAS

Agent runtimes like ZeroClaw and OpenClaw proved that AI agents can be fast, tiny, and deployable anywhere. But they were not built for environments where every action must be traceable, policy-constrained, and verifiable.

VERITAS does not replace them. It wraps them with trust.

Linux Kernel        →  ZeroClaw / OpenClaw    (fast, minimal, runs anywhere)
Red Hat Enterprise  →  VERITAS                (trusted, governed, auditable)

Architecture

┌─────────────────────────────────────────────────────┐
│              Application / Agent Code               │
│         (LangGraph, CrewAI, OpenClaw, etc.)         │
├─────────────────────────────────────────────────────┤
│                    VERITAS                           │
│   Policy Engine │ Audit Trail │ Verifier │ Caps     │
├─────────────────────────────────────────────────────┤
│              Agent Runtime Kernel                    │
│            (ZeroClaw or equivalent)                  │
└─────────────────────────────────────────────────────┘

Execution Model

Every agent action follows the same deterministic pipeline — no exceptions, no shortcuts:

State → Policy → Capability → Audit → Verify → Next State

Trust Boundary

Trusted	Untrusted
Runtime core	LLM
Policy engine	Tools
Audit engine	Input data
Verifier	External environment

Core Components

Crate	Purpose	Tests
veritas-contracts	Shared types, traits, error types	15
veritas-core	Deterministic executor pipeline	6
veritas-policy	TOML deny-by-default policy engine	8
veritas-audit	SHA-256 hash-chained audit trail	6
veritas-verify	JSON Schema + semantic rule verification	10
veritas-ref-healthcare	Healthcare reference runtime (5 scenarios)	13

Healthcare Demo Scenarios

#	Scenario	What it demonstrates
1	Drug Interaction Checker	Policy Allow flow, output schema verification
2	Clinical Note Summarizer	PII detection via custom verifier rule
3	Patient Data Query	Capability-based access control, consent enforcement
4	Multi-Agent Clinical Pipeline	4-agent chain with independent audit trails
5	Prior Authorization Workflow	RequireApproval lifecycle with physician approval

Run individually:

cargo run -p demo -- drug-interaction
cargo run -p demo -- note-summarizer
cargo run -p demo -- patient-query
cargo run -p demo -- clinical-pipeline
cargo run -p demo -- prior-auth

Design Principles

1. Control over autonomy
2. Evidence over intelligence
3. Determinism over emergence
4. Deny by default
5. Capability-based security
6. Minimal trusted computing base
7. Auditability by design
8. Verifiable execution
9. Human override always possible
10. Data-model independence

Lightweight by conviction. Governance must not be the reason agents become slow, heavy, or hard to build.

Project Structure

crates/
  veritas-contracts/       # Shared types, traits, error types
  veritas-core/            # Deterministic executor pipeline
  veritas-policy/          # TOML deny-by-default policy engine
  veritas-audit/           # SHA-256 hash-chained audit trail
  veritas-verify/          # JSON Schema + semantic rule verification
  veritas-ref-healthcare/  # Healthcare reference runtime (5 scenarios)
demo/                      # CLI demo runner (clap)
tui/                       # Interactive TUI demo (ratatui)
docs/
  whitepaper/              # Whitepaper v0.3 (EN, ZH, JA, FR)
  yellowpaper/             # Yellow Paper v0.1 (EN)

Documentation

Document	Description
Whitepaper v0.3	Vision, design philosophy, system architecture
Yellow Paper v0.1	Formal execution semantics and specifications
docs/	Full documentation index

Whitepaper Translations

Language	Link
English	WHITEPAPER.en.md
简体中文	WHITEPAPER.zh.md
日本語	WHITEPAPER.ja.md
Français	WHITEPAPER.fr.md

Contributing

VERITAS is open source. Community contributions — including new translations — are welcome. See CONTRIBUTING.md for details.

License

Licensed under Apache License 2.0. See LICENSE for details.