OpenGradient Developer Cookbook

A curated collection of production-ready code snippets, boilerplates, and patterns for building applications on the OpenGradient platform — the verifiable AI inference network.

This cookbook is the foundation for the OG Helper Telegram Bot which serves these examples to developers on demand.

---

What is OpenGradient?

OpenGradient provides verifiable AI inference — every LLM call runs inside a Trusted Execution Environment (TEE) and settles payment on Base Sepolia via the x402 protocol. Each inference returns a transaction_hash proving:

• Which exact model was used
• What the exact input and output were
• That the computation happened in an isolated secure enclave

---

Repository Structure

OpenGradient-Developer-Cookbook/
├── snippets/                    # Standalone runnable examples
│   ├── 01_llm_completion_basic.py   # Single completion + proof
│   ├── 02_llm_chat_with_tools.py    # Tool calling / function calling
│   ├── 03_llm_streaming.py          # Streaming completions
│   ├── 04_settlement_modes.py       # PRIVATE / INDIVIDUAL / BATCH settlement
│   ├── 05_model_hub_upload.py       # Upload ONNX model to Model Hub
│   ├── 06_check_opg_balance.py      # Wallet inspector (ETH + OPG + Permit2)
│   ├── 07_memsync_persistent.py     # Persistent AI memory via MemSync
│   ├── 08_ml_inference_alpha.py     # On-chain ONNX inference (alpha testnet)
│   ├── 09_workflow_alpha.py         # On-chain ML workflow (alpha testnet)
│   ├── 10_permit2_approval.py       # Manual Permit2 OPG allowance approval
│   └── 11_langchain_agent.py        # LangGraph ReAct agent (og.agents adapter)
├── utils/
│   └── client.py                    # Singleton accessors: get_llm, get_alpha, get_hub
├── .env.example                     # Environment variable template
├── requirements.txt                 # Python dependencies
└── README.md

---

Quick Start

Prerequisites

• Python 3.11+ (required by opengradient >= 0.6.0)
• A testnet wallet with OG_PRIVATE_KEY (Base Sepolia)
• $OPG test tokens from the faucet
• ETH for gas from Base Sepolia faucet

Installation

git clone https://github.com/Floyd11/OpenGradient-Developer-Cookbook.git
cd OpenGradient-Developer-Cookbook

python3 -m venv .venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate

pip install -r requirements.txt

Configuration

cp .env.example .env
# Edit .env — set OG_PRIVATE_KEY (testnet wallet only, never mainnet!)

Verify your wallet

# Check ETH balance, $OPG balance, and Permit2 allowance
python snippets/06_check_opg_balance.py

Run your first inference

python snippets/01_llm_completion_basic.py

---

Snippets Overview

#	Snippet	What it demonstrates
01	01_llm_completion_basic.py	Single verifiable completion + transaction_hash proof
02	02_llm_chat_with_tools.py	Tool calling / function calling with TEE-verified LLM
03	03_llm_streaming.py	Real-time streaming completions
04	04_settlement_modes.py	PRIVATE, INDIVIDUAL_FULL, BATCH_HASHED settlement
05	05_model_hub_upload.py	Upload + manage ONNX models on Model Hub
06	06_check_opg_balance.py	Inspect ETH / $OPG / Permit2 allowance
07	07_memsync_persistent.py	Persistent AI memory via MemSync API
08	08_ml_inference_alpha.py	VANILLA / ZKML / TEE on-chain ONNX inference ⚠️ alpha
09	09_workflow_alpha.py	Deploy + run on-chain ML workflow ⚠️ alpha
10	10_permit2_approval.py	Manual Permit2 $OPG allowance management
11	11_langchain_agent.py	LangGraph ReAct agent via og.agents.langchain_adapter()

---

Supported Models (SDK 0.9.3+)

Provider	Model enum	Notes
OpenAI	og.TEE_LLM.GPT_5	Default recommended
OpenAI	og.TEE_LLM.GPT_5_MINI	Faster / cheaper
OpenAI	og.TEE_LLM.GPT_5_2	Latest GPT-5 variant
OpenAI	og.TEE_LLM.GPT_4_1_2025_04_14	GPT-4.1
OpenAI	og.TEE_LLM.O4_MINI	Reasoning model
Anthropic	og.TEE_LLM.CLAUDE_SONNET_4_6	Recommended Claude
Anthropic	og.TEE_LLM.CLAUDE_SONNET_4_5	Claude Sonnet 4.5
Anthropic	og.TEE_LLM.CLAUDE_OPUS_4_6	Most capable Claude
Anthropic	og.TEE_LLM.CLAUDE_OPUS_4_5	Claude Opus 4.5
Anthropic	og.TEE_LLM.CLAUDE_HAIKU_4_5	Fast / lightweight
Google	og.TEE_LLM.GEMINI_3_PRO	Latest Gemini
Google	og.TEE_LLM.GEMINI_3_FLASH	Fast Gemini 3
Google	og.TEE_LLM.GEMINI_2_5_PRO	Gemini 2.5 Pro
Google	og.TEE_LLM.GEMINI_2_5_FLASH	Gemini 2.5 Flash
Google	og.TEE_LLM.GEMINI_2_5_FLASH_LITE	Lightweight Flash
xAI	og.TEE_LLM.GROK_4	Grok 4
xAI	og.TEE_LLM.GROK_4_FAST	Grok 4 Fast
xAI	og.TEE_LLM.GROK_4_1_FAST	Grok 4.1 Fast
xAI	og.TEE_LLM.GROK_4_1_FAST_NON_REASONING	Grok 4.1 non-reasoning

---

Architecture: utils/client.py

All snippets share a single utils/client.py that provides lazy-initialized singleton accessors:

from utils.client import get_llm, get_alpha, get_hub

# LLM inference (verifiable via x402 Gateway) — needs OG_PRIVATE_KEY
llm = get_llm()

# On-chain ML inference / workflows (alpha testnet) — needs OG_PRIVATE_KEY
alpha = get_alpha()

# Model Hub management — needs OG_EMAIL + OG_PASSWORD
hub = get_hub()

Note: og.Client was removed from the SDK in version 0.6.0. Use the dedicated classes above (og.LLM, og.Alpha, og.ModelHub) instead.

---

Environment Variables

See .env.example for the full template.

Variable	Required	Description
OG_PRIVATE_KEY	✅ Always	Testnet wallet private key (Base Sepolia — never use mainnet!)
OG_EMAIL	Model Hub only	OpenGradient Model Hub account email
OG_PASSWORD	Model Hub only	OpenGradient Model Hub account password
RPC_URL	Optional	Base Sepolia RPC (default: https://sepolia.base.org)
CHAIN_ID	Optional	Chain ID (default: 84532)
OPG_TOKEN_CONTRACT	Optional	OPG ERC-20 address (default: 0x240b09731D96979f50B2C649C9CE10FcF9C7987F)
LOG_LEVEL	Optional	Logging verbosity: DEBUG, INFO, WARNING (default: INFO)

---

Security Notes

• 🔐 Testnet only: OG_PRIVATE_KEY must be a dedicated testnet wallet. Never use a mainnet private key.
• 🚫 Never commit .env: it is in .gitignore by default.
• ✅ Permit2 pattern: the x402 Gateway uses Permit2 for gasless $OPG spending approval — no raw approve() calls needed.

---

Resources

• 📖 OpenGradient Docs
• 🐍 OpenGradient Python SDK on PyPI
• 🤖 OG Agent Starter
• 💬 Discord Community
• 🚧 OG Explorer (testnet)
• 🚰 OPG Faucet