OpenGradient Developer Cookbook 🍳

The ultimate collection of boilerplates, snippets, and templates for building trustless, verifiable AI applications using the OpenGradient SDK.

---

🏗️ Architecture Overview

Your App
   │
   ▼
og.LLM (x402 Gateway)
   │  Permit2 $OPG payment on Base Sepolia
   ▼
TEE Node (Intel TDX Confidential Compute)
   │  Hardware-attested code execution
   ▼
OpenAI / Anthropic / Google / xAI API
   │
   ▼
Base Sepolia  ←── $OPG payment + TEE attestation proof
   │
   ▼
OpenGradient Network  ←── Proof settlement (Merkle / Full / Private)

Every LLM call through OpenGradient returns a payment_hash — a cryptographic on-chain proof that a specific prompt was processed inside a Trusted Execution Environment. This makes AI decisions auditable, verifiable, and trustless.

---

📚 Table of Contents

Snippets

#	File	What it demonstrates
01	01_llm_completion_basic.py	Simple verifiable LLM text completion
02	02_llm_chat_with_tools.py	Multi-turn chat with function/tool calling
03	03_llm_streaming.py	Real-time streaming chat output
04	04_settlement_modes.py	PRIVATE / INDIVIDUAL_FULL / BATCH_HASHED modes
05	05_model_hub_management.py	Model Hub: create → version → upload → list
06	06_check_opg_balance.py	Wallet inspector: OPG + ETH balance on Base Sepolia
07	07_memsync_personalized_bot.py	MemSync: store → search → profile → chatbot
08	08_ml_inference_alpha.py	On-chain ML inference (VANILLA / ZKML / TEE)
09	09_ml_workflow_deploy.py	Deploy + read automated ML volatility workflow
10	10_permit2_approval.py	OPG Permit2 token approval flow
11	11_langchain_agent.py	LangChain agent with OG verifiable LLM backend

Boilerplates

Boilerplate	Description
fastapi-verifiable-backend	Production FastAPI server with verifiable AI inference endpoints
antigravity-ai-agent	Autonomous agent loop with on-chain proof trail
defi-risk-analyzer	DeFi smart contract auditor with TEE-verified analysis
memsync-chatbot	Personalized chatbot with persistent MemSync memory
digital-twins-tracker	Twin.fun share price tracker + AI commentary

---

🚀 Getting Started

Prerequisites

• Python 3.10+
• A testnet-only Ethereum wallet (never use mainnet funds)
• Test $OPG tokens on Base Sepolia — get them at faucet.opengradient.ai
• OpenGradient Model Hub account — register at hub.opengradient.ai
• MemSync API key (optional, for snippets 07+) — register at app.memsync.ai

Installation

# 1. Clone the repo
git clone https://github.com/your-org/OpenGradient-Cookbook.git
cd OpenGradient-Cookbook

# 2. Create a virtual environment
python -m venv .venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate

# 3. Install dependencies
pip install -r requirements.txt

# 4. Configure environment variables
cp .env.example .env
# Edit .env with your testnet private key and credentials

Environment Setup

# .env — fill in your values
OG_PRIVATE_KEY=0xYOUR_TESTNET_PRIVATE_KEY
OG_EMAIL=your@email.com
OG_PASSWORD=your_model_hub_password
MEMSYNC_API_KEY=your_memsync_key

⚠️ SECURITY: The private key in .env must be a testnet-only wallet. Never put mainnet funds in a wallet used for development. Your .env file is in .gitignore and will never be committed.

---

📝 Snippets

01 — LLM Completion Basic {#01}

Simple verifiable text completion. The "Hello World" of OpenGradient.

python snippets/01_llm_completion_basic.py

02 — LLM Chat with Tools {#02}

Multi-turn chat with function/tool calling. Build agents that call external APIs.

python snippets/02_llm_chat_with_tools.py

03 — LLM Streaming {#03}

Real-time streaming output for terminal apps and web UIs.

python snippets/03_llm_streaming.py

04 — Settlement Modes {#04}

Compare PRIVATE / INDIVIDUAL_FULL / BATCH_HASHED — choose the right privacy/cost tradeoff.

python snippets/04_settlement_modes.py

05 — Model Hub Management {#05}

Create a model repo, version it, upload an ONNX file, and list files.

python snippets/05_model_hub_management.py

06 — Check OPG Balance {#06}

Inspect your wallet's ETH and $OPG balance before running inference.

python snippets/06_check_opg_balance.py

07 — MemSync Personalized Bot {#07}

Store conversations → search memories → retrieve user profile using MemSync.

python snippets/07_memsync_personalized_bot.py

08 — ML Inference Alpha {#08}

Run on-chain ML model inference in VANILLA, ZKML, and TEE modes (alpha testnet).

python snippets/08_ml_inference_alpha.py

09 — ML Workflow Deploy {#09}

Deploy an automated ETH volatility prediction workflow with oracle data feeds.

python snippets/09_ml_workflow_deploy.py

10 — Permit2 Approval {#10}

Approve $OPG tokens via Permit2 — required once before any LLM inference.

python snippets/10_permit2_approval.py

11 — LangChain Agent {#11}

Use OpenGradient's verifiable LLM as a drop-in LangChain backend.

python snippets/11_langchain_agent.py

---

🏗️ Boilerplates

FastAPI Verifiable Backend

A production-ready REST API server with /infer, /chat, and /stream endpoints. Returns verifiable payment_hash with every AI response.

Antigravity AI Agent

Fully autonomous agent that breaks tasks into steps, executes them with verifiable LLM inference, and logs a cryptographic proof trail to agent_log.jsonl.

DeFi Risk Analyzer

Analyzes Solidity smart contracts and loan applications using TEE-verified LLM. Every analysis includes a payment_hash for regulatory audit compliance.

MemSync Chatbot

A personalized chatbot that remembers users across sessions using MemSync's semantic memory layer combined with OpenGradient's verifiable LLM.

Digital Twins Tracker

Track Twin.fun share prices, check your holdings, and get AI-generated market commentary — all with verifiable on-chain proof of the AI analysis.

---

🎯 Use Cases

Domain	Use Case	Key Feature
DeFi	Smart contract audits, trading agent decisions	INDIVIDUAL_FULL settlement — proof of every prompt
Finance	Loan risk assessment, fraud detection	TEE verification — hardware-attested privacy
Healthcare	Patient data analysis, clinical decision support	TEE confidential compute — data never exposed
Enterprise AI	Content moderation, compliance logging	Verifiable audit trails for regulatory requirements
AI Agents	Autonomous agents with provable actions	payment_hash per step — full reasoning transparency
Consumer Apps	Personalized assistants, tutors, CRM	MemSync — persistent memory across sessions

---

🤝 Contributing

Contributions are welcome! Here's how:

1. Fork the repository
2. Create a feature branch: git checkout -b feature/my-snippet
3. Follow the coding standards:
   • Python 3.10+ with full type hints
   • Google-style docstrings on all modules and functions
   • logging for operational messages, print() for user-facing output
   • All secrets via os.getenv() — never hardcoded
4. Test your snippet: python snippets/your_snippet.py
5. Open a Pull Request with a clear description

Adding a New Snippet

• Place it in snippets/ with the next number prefix (e.g. 12_my_feature.py)
• Add it to the Table of Contents in this README
• Use utils/client.py for SDK initialization

Adding a New Boilerplate

• Create a new folder in boilerplates/
• Include main.py (or equivalent), requirements.txt, and README.md
• The README must include: what it does, how to run, example output

---

📖 Resources

• OpenGradient Docs
• Python SDK Reference
• x402 Gateway Overview
• MemSync Guide
• Model Hub
• Block Explorer
• Testnet Faucet
• Discord Community

---

📄 License

MIT — see LICENSE for details.