Skip to content
Vera-ProtocalPublic

About

The verification layer for AI agents that independently checking agent work, evidence, and outcomes before they can be trusted.

Resources

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

Latest commit

 

History

111 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

VeraOS

The verification layer for AI agents.
Verify before you trust.

CI Tests TypeScript Stellar License: MIT

Autonomous AI agents are increasingly entrusted with high-stakes actions: settling payments, dispatching bounties, deploying liquidity, and executing Soroban smart contracts on Stellar. However, agents self-report their success. When an agent reports "Payment of 5 USDC executed successfully", systems have historically trusted that claim at face value.

VeraOS replaces blind trust with automated, deterministic verification. Sitting between task orchestrators and autonomous agents, VeraOS independently extracts requirements, pulls authoritative ground-truth evidence directly from primary sources (such as the Stellar blockchain ledger), runs deterministic verification checks, and renders a structured verdict with actionable remediation directives.


1. Product Overview

  • Positioning: The verification layer for AI agents.
  • Core Mantra: Verify before you trust.
  • Problem: Large language models and autonomous agent frameworks suffer from the self-reporting fallacy. An agent may hallucinate a transaction hash, underpay due to slippage or parameter error, send to the wrong recipient, or report success on a reverted transaction.
  • Solution: VeraOS never treats an agent's words as evidence. It extracts the claimed hypothesis and independently proves or disproves it against authoritative blockchain RPC and cryptographic evidence.

2. Core Architecture

┌────────────────────────────────────────────────────────────────────────┐
│                        User & Operator Ingress                         │
│  - Web Dashboard (React 19 + Tailwind CSS)                             │
│  - Telegram Bot (@VeraOSBot via long-polling runner or webhook)        │
│  - REST API Client / Agent Orchestrator (POST /v1/verify)              │
└───────────────────────────────────┬────────────────────────────────────┘
                                    │
                                    ▼
┌────────────────────────────────────────────────────────────────────────┐
│                      VeraOS Verification Pipeline                      │
│                                                                        │
│   1. Requirement Extractor                                             │
│      Task Prompt ──► [ Invariant Rules & Expected Bounds ]             │
│                                                                        │
│   2. Worker Claim Extractor                                            │
│      Worker Output ──► [ Extracted Claims & Assertions ]               │
│                                                                        │
│   3. Deterministic Check Engine                                        │
│      Triangulates claims against independent evidence providers:       │
│      ├── Stellar RPC Provider (Live Testnet Ledger & Horizon)          │
│      ├── Deterministic Kernel (Cardinality, bounds, structure)         │
│      └── Worker Output Provider (Baseline trace audit)                 │
│                                                                        │
│   4. Mathematical Verdict Synthesizer                                  │
│      Checks Map ──► VERIFIED | FAILED | PARTIAL | UNVERIFIABLE         │
│                                                                        │
│   5. Remediation Directive Engine                                      │
│      Failures ──► Structured Correction Directives (Amount, Target)    │
└───────────────────────────────────┬────────────────────────────────────┘
                                    │
                                    ▼
┌────────────────────────────────────────────────────────────────────────┐
│                           Persistence & Audit                          │
│  - Verification Repository (Attempts, Checks, Evidence, Directives)    │
│  - Public Explorer Proofs (Stellar.Expert Testnet Links)               │
└────────────────────────────────────────────────────────────────────────┘

3. Real Stellar Testnet Verification

VeraOS connects directly to Stellar Soroban RPC (https://soroban-testnet.stellar.org) and Stellar Horizon (https://horizon-testnet.stellar.org), decoding the signed transaction envelope XDR via @stellar/stellar-sdk.

Critical Benchmark Scenario: Deceptive Worker Detection

  1. Task: "Send 5 USDC to GCEYAUYCI3WTE5GOD7CDLRJQPATQCLHMXY4Q3CEQ64RP5SVDWPFF5L2L."
  2. Worker Claim: "Payment completed. I sent 5 USDC to GCEYAU... via transaction 108822f67b10e3ad38db576d60712939c1bdbe372c9d4729928d613605682759."
  3. VeraOS Result:
    • Queries Stellar RPC for hash 108822f6....
    • Discovers actual ledger payment operation transferred only 0.50 USDC.
    • Verdict: FAILED.
    • Structured Delta Check:
      Expected: 5 USDC
      Observed: 0.5 USDC
      Difference: -4.5 USDC
      Payment mismatch: Expected 5 USDC, but observed receipt shows 0.5 USDC (Deficit: 4.50 USDC).
      

Live Testnet Verification Artifacts

Role Account / Transaction Hash Explorer Link
Recipient GCEYAUYCI3WTE5GOD7CDLRJQPATQCLHMXY4Q3CEQ64RP5SVDWPFF5L2L View on Stellar.Expert
Deceptive Tx (0.5 USDC) 108822f67b10e3ad38db576d60712939c1bdbe372c9d4729928d613605682759 View 0.5 USDC Tx
Corrected Tx (5.0 USDC) 62256096f306726197208231b00e422628b0bb83e104dabed9a74da5186afbaf View 5.0 USDC Tx

4. Quick Start

1. Installation

git clone https://github.com/k-deejah/VeraOS.git
cd VeraOS
npm install
cp .env.example .env

2. Run Locally

# Start Vite development server + integrated API on http://localhost:5173
npm run dev

# Start standalone API server on http://localhost:3001
npm run server

# Start Telegram bot long-polling runner
npm run bot

3. Run Automated Tests

# Runs complete 28-test suite (engine, stellar, stellarRpc, telegram)
npm test

# Run linter and typecheck
npm run lint
npx tsc -b

4. Docker Deployment

# Run via Docker Compose
docker compose up --build

5. Telegram Bot Interface

VeraOS includes a dedicated, production-grade Telegram bot supporting both long-polling (getUpdates) and webhook modes.

Commands

  • /start — Welcome message and command guide.
  • /verify — Start interactive verification or inline evaluation (/verify <task> | <output>).
  • /status <id> — Check live status card of a verification.
  • /evidence <id> — View onchain evidence and Stellar.Expert explorer links.
  • /correct <id> [note] — Issue self-correction directive.
  • /resubmit <id> [tx] — Re-evaluate with corrected transaction hash.

See docs/TELEGRAM_BOT.md for full configuration and security documentation.


6. REST API Reference

Execute Verification: POST /v1/verify

curl -X POST http://localhost:5173/v1/verify \
  -H "Content-Type: application/json" \
  -d '{
    "task": "Send 5 USDC to GCEYAUYCI3WTE5GOD7CDLRJQPATQCLHMXY4Q3CEQ64RP5SVDWPFF5L2L.",
    "worker": {
      "id": "agent-alpha-09",
      "output": "Payment completed via tx 108822f67b10e3ad38db576d60712939c1bdbe372c9d4729928d613605682759."
    }
  }'

See docs/API_REFERENCE.md for complete endpoint schemas.


7. Product & Technical Roadmap

Our prioritized technical milestones focus on enterprise scalability and zero-trust verification:

  • Cloudflare D1 Ledger Storage: SQLite database schema with persistent user sessions and audit logs.
  • Stellar Horizon RPC Corroboration: Live onchain payment and contract invocation proofs.
  • Automated Remediation: Actionable correction loops for failed agent outputs.

Check out docs/CONTRIBUTOR_ROADMAP.md for full technical milestone details.


8. Documentation Index


9. Security & Governance

  • Security Policy: See SECURITY.md for vulnerability disclosure and reporting.
  • Contributing: See CONTRIBUTING.md for development workflows and PR templates.
  • License: Released under the MIT License.
  • Maintainers: Maintained by @k-deejah and the VeraOS Open Source Contributors.

About

The verification layer for AI agents that independently checking agent work, evidence, and outcomes before they can be trusted.

Resources

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages