# Halo

> A permissionless, peer-to-peer marketplace for verifiable AI inference. Access 140+ AI models — including GPT-5, Claude, Gemini, DeepSeek, Qwen and Kimi — with no account and no subscription, paying per prompt in USDC on the Base network.

Halo (built by Warden) is "BitTorrent for AI": instead of prompts funnelling through a handful of data centres, a global mesh of independent operators serves AI models peer-to-peer and can cryptographically prove they ran the real model (Statistical Proof of Execution, SPEX). Anyone can consume inference or run an operator to earn USDC. Prompts stay private — encrypted to the operator, with an optional hardware-enclave (TEE) confidential mode — and there are no API keys, accounts, or rate limits. Payment is per prompt in USDC on Base; because independent operators compete for each request, prices are typically among the lowest available for a given model. The site is available in English (default) plus Chinese, Vietnamese, Ukrainian, Russian and Spanish (under /zh/, /vi/, /uk/, /ru/, /es/).

## Start here
- [The Halo Manifesto (home)](https://runhalo.xyz/): What Halo is and why it exists.
- [Verification — Statistical Proof of Execution](https://runhalo.xyz/#spex): How Halo proves an operator ran the real model.
- [Use Halo (the app)](https://app.runhalo.xyz): The live Halo application.

## Guides
- [Consume inference with your agent](https://runhalo.xyz/guides/agents-consume): Let an AI agent set up Halo by chatting — pick a model, choose confidential mode, fund a wallet, and it has pay-per-prompt inference with no API keys.
- [Consume inference from a local endpoint (CLI)](https://runhalo.xyz/guides/developers-consume): Use the halo CLI to run a local OpenAI-compatible endpoint that pays per request from your wallet — deposit once, bill actual token usage, with spend guards.
- [Use Halo on the web](https://runhalo.xyz/guides/use-halo-on-the-web): The easiest way to use Halo: the web app. Sign in, top up with card or crypto, pick a model, and start prompting — with optional confidential (TEE) inference.
- [What to serve as an operator](https://runhalo.xyz/guides/what-to-serve): Choosing what your Halo operator serves — a local GPU model (Ollama/LM Studio), a provider API key (OpenAI, OpenRouter…), or a hosted model — and which models earn.
- [Operate and earn with your agent](https://runhalo.xyz/guides/agents-operate): Have your AI agent serve inference on Halo and earn USDC — set it up by chatting: choose what to serve, pick a pricing mode, and let it run. No pre-funding.
- [Run an operator (CLI)](https://runhalo.xyz/guides/developers-operate): Use the halo CLI to serve inference from a provider, hosted model, or local GPU and earn USDC on Base — provider slugs, pricing, settlement, and always-on setup.
- [Fund your account: the Wallet & Vault](https://runhalo.xyz/guides/the-halo-vault): Your Halo balance lives in an on-chain vault on Base. Add funds with a card or crypto, read Free vs Reserved, and withdraw what you haven't spent.
- [Chat mode vs Agent mode (and effort)](https://runhalo.xyz/guides/chat-vs-agent-mode): Halo has two modes — plain Chat and the full Agent loop — plus effort tiers (Auto, Quick, Standard, Deep) and run modes. Learn when to use each.
- [Link your agent](https://runhalo.xyz/guides/agents-league-and-stats): Link your agent's operator and consumer activity to your Halo dashboard to see your stats and earnings, and climb the League — all by chatting with your agent.
- [Serve local models on your GPU (Llama, Ollama, LM Studio)](https://runhalo.xyz/guides/serve-local-models): Turn an idle GPU into Halo earnings — serve Llama, Qwen, Gemma and other open models via Ollama or LM Studio, price them flat, and go always-on. No API key needed.
- [Generate images](https://runhalo.xyz/guides/generate-images): Create images from a prompt on Halo: pick an image model, write a prompt, pay per image in USDC. Peer-to-peer, end-to-end encrypted, stored nowhere.
- [Edit images](https://runhalo.xyz/guides/edit-images): Upload a picture, describe the change, get the edited result — peer-to-peer and end-to-end encrypted. EXIF is stripped before upload; nothing is stored.
- [Memory & context](https://runhalo.xyz/guides/memory-and-context): Halo remembers facts across chats — stored locally in your browser, never on a server. Learn how to save, tag, pin, and scope memory as context for the agent.
- [Update Halo with your agent](https://runhalo.xyz/guides/agents-update): Keeping Halo current is one sentence: tell your agent to update Halo to the latest version. It refreshes the CLI, restarts services, and verifies — wallet untouched.
- [Operator pricing & earnings](https://runhalo.xyz/guides/operator-pricing-and-earnings): How to price a Halo operator — margin vs flat — how the 10% protocol fee and USDC settlement on Base work, and how to maximize what you net per request.
- [Keep your operator online (always-on & reliability)](https://runhalo.xyz/guides/operator-reliability): Run your Halo operator as a persistent service so it stays online, earns more, and builds reputation — install it as a service, watch the logs, and survive restarts.
- [Enable your agent session](https://runhalo.xyz/guides/tools-and-x402): Give the agent a session sub-wallet so it can use paid tools — web search, scraping, price data, or custom x402 endpoints — autonomously, and cap the spend.
- [Update your operator (without losing your identity)](https://runhalo.xyz/guides/operator-update): Update a running Halo operator safely: refresh the CLI, restart the service, verify with doctor — wallet, reputation, and earnings all carry over.
- [Private prompts: encryption & confidential compute](https://runhalo.xyz/guides/privacy-and-confidential-compute): Every Halo prompt is end-to-end encrypted. Confidential mode seals it to a hardware enclave the operator can't read, with on-device attestation you can verify.
- [Pair with the dashboard (CLI)](https://runhalo.xyz/guides/pair-with-the-dashboard): Link a halo CLI wallet — operator or consumer — to a Halo dashboard account to monitor your stats on the web and join the League. Here's the pairing flow.
- [Export an answer to PDF](https://runhalo.xyz/guides/export-to-pdf): Turn any Halo answer into a clean, branded Research Report PDF — with its charts, tables, and sources — generated right in your browser, at no extra cost.
- [Update the halo CLI](https://runhalo.xyz/guides/update-the-cli): Get the latest halo CLI in two commands — remove the old link, re-run the installer, verify with doctor. Your wallet, config, and earnings are untouched.

## AI models on Halo
- [Use DeepSeek without a subscription — DeepSeek V3.2, V4 & R1 on Halo](https://runhalo.xyz/models/deepseek): Run DeepSeek with no subscription and no account — pay per prompt in USDC on Base. Verifiable, private, and reachable from anywhere. No Chinese phone number needed.
- [Use Qwen without a subscription — Alibaba's Qwen3 on Halo](https://runhalo.xyz/models/qwen): Run Alibaba's Qwen (Qwen3.5, Qwen3.6) with no subscription and no account — pay per prompt in USDC on Base. Verifiable, private, and reachable from anywhere.
- [Use Kimi (Moonshot) without a subscription — Kimi K2 on Halo](https://runhalo.xyz/models/kimi): Run Moonshot's Kimi (K2.5, K2.6) with no subscription and no account — pay per prompt in USDC on Base. Verifiable, private, and reachable from anywhere.
- [Use GLM (Zhipu / Z.ai) without a subscription — GLM-5 on Halo](https://runhalo.xyz/models/glm): Run Zhipu's GLM (GLM-4.7, GLM-5) with no subscription and no account — pay per prompt in USDC on Base. Verifiable, private, and reachable from anywhere.
- [Use MiniMax without a subscription — MiniMax M2 on Halo](https://runhalo.xyz/models/minimax): Run MiniMax (M2, M3) with no subscription and no account — pay per prompt in USDC on Base. Verifiable, private, and reachable from anywhere.
- [Use GPT-5 without a ChatGPT subscription — pay per prompt on Halo](https://runhalo.xyz/models/gpt-5): Reach GPT-5, the model behind ChatGPT, with no OpenAI account and no monthly plan. Pay per prompt in USDC on Base — verifiable, private, available anywhere.
- [Use Claude without a subscription — Opus, Sonnet & Haiku on Halo](https://runhalo.xyz/models/claude): Reach Claude with no Anthropic account and no monthly plan. Pay per prompt in USDC on Base — verifiable, private, and reachable from anywhere.
- [Use Gemini without a subscription — pay per prompt on Halo](https://runhalo.xyz/models/gemini): Reach Google's Gemini with no Google account tie-in and no monthly plan. Pay per prompt in USDC on Base — verifiable, private, reachable anywhere.
- [Run Llama without hosting it — pay per prompt on Halo](https://runhalo.xyz/models/llama): Use Meta's Llama models without renting a GPU or hosting anything. Independent operators serve them on Halo — pay per prompt in USDC on Base, no account.
- [The best Chinese ChatGPT alternatives (DeepSeek, Qwen, Kimi, GLM) — no subscription](https://runhalo.xyz/models/chinese-chatgpt-alternatives): The top Chinese AI models — DeepSeek, Qwen, Kimi, GLM, MiniMax — and how to use any of them on Halo with no subscription, no account, pay per prompt in USDC.
- [Use Mistral models pay-per-prompt — no subscription, no account](https://runhalo.xyz/models/mistral): Run Mistral's open models on Halo without a provider account or a monthly plan. Pay per prompt in USDC on Base — verifiable, private, reachable anywhere.
- [DeepSeek vs ChatGPT — how they compare, and how to run either without a subscription](https://runhalo.xyz/models/deepseek-vs-chatgpt): DeepSeek vs ChatGPT compared on cost, reasoning, privacy and access — plus how to use either on Halo with no subscription, no account, pay per prompt in USDC.
- [Claude vs ChatGPT — how they compare, and how to run either without a subscription](https://runhalo.xyz/models/claude-vs-chatgpt): Claude vs ChatGPT on reasoning, writing, context and cost — plus how to run either on Halo with no account and no monthly plan, paying per prompt in USDC.
- [Gemini vs ChatGPT — how they compare, and how to run either without a subscription](https://runhalo.xyz/models/gemini-vs-chatgpt): Gemini vs ChatGPT on context length, multimodal work, reasoning and cost — plus how to run either on Halo with no account, paying per prompt in USDC.
- [Qwen vs DeepSeek — how the two top Chinese models compare](https://runhalo.xyz/models/qwen-vs-deepseek): Qwen vs DeepSeek on reasoning, coding, multilingual work and cost — plus how to run either on Halo with no account, no subscription, paying per prompt.
- [Access ChatGPT & Claude in Russia — no account, no subscription](https://runhalo.xyz/models/chatgpt-claude-in-russia): ChatGPT and Claude aren't available in Russia. Use GPT-5, Claude and 140+ models on Halo instead — pay per prompt in USDC, no account, no subscription, no foreign card.
- [Access ChatGPT & Claude in China — no account, no subscription](https://runhalo.xyz/models/chatgpt-claude-in-china): ChatGPT and Claude aren't available in mainland China. Use GPT-5, Claude and 140+ models on Halo — pay per prompt in USDC, no account, no subscription, no Chinese number.
- [Access ChatGPT & Claude in Hong Kong — no account, no subscription](https://runhalo.xyz/models/chatgpt-claude-in-hong-kong): ChatGPT and Claude aren't offered in Hong Kong. Use GPT-5, Claude and 140+ models on Halo — pay per prompt in USDC on Base, no account, no subscription, no VPN sign-up.

## Blog
- [One account for you and your agents](https://runhalo.xyz/blog/one-account-everywhere): Sign in with a social account, an email, a passkey, or a wallet. One identity and one balance across the app and every agent you run.
- [The first HALO buyback: 219,944 tokens burned](https://runhalo.xyz/blog/first-halo-buyback-and-burn): Cycle 1 of the HALO buyback is settled on Base: 219,944 HALO burned permanently, 513,203 to stakers. Every step links to a public transaction.
- [$50,000 in HALO for volume and net buyers](https://runhalo.xyz/blog/halo-trading-campaign): A 14-day trading campaign on the HALO/VIRTUAL pool on Aerodrome. $50,000 in HALO: half pro rata to traded volume, half to net buying volume.
- [The HALO whitepaper is here](https://runhalo.xyz/blog/halo-whitepaper): The full design of the open inference economy — SPEX verification, the fee-to-buyback loop, staking, emissions, genesis, governance. And no venture allocation.
- [Image generation on Halo: peer-to-peer, end-to-end encrypted, stored nowhere](https://runhalo.xyz/blog/image-generation-on-halo): Generate and edit images over Halo's P2P rail. No central servers, no cloud storage — encrypted bytes straight between peers, and you pay per image delivered.
- [Access every AI model without a subscription — from anywhere in the world](https://runhalo.xyz/blog/every-ai-model-no-subscription): Reach every frontier model — GPT, Claude, Gemini, DeepSeek and more — in one place. No subscriptions, no accounts, pay per prompt in USDC, from anywhere.
- [Introducing Halo: intelligence that belongs to everyone](https://runhalo.xyz/blog/introducing-halo): Halo is a permissionless, peer-to-peer marketplace for AI inference — request any model, serve one, and settle in USDC on Base. Our public alpha is live.
- [How to monetize your AI API keys — resell inference per call, paid in stablecoins](https://runhalo.xyz/blog/monetize-your-ai-api-keys): Turn spare AI API access, unused credits, or an idle GPU into income — resell inference per call on Halo, permissionlessly, and get paid instantly in USDC.
- [What is verifiable AI inference — and why it matters](https://runhalo.xyz/blog/what-is-verifiable-ai-inference): When you call an AI model, you trust the vendor ran what you paid for. Verifiable inference replaces that trust with proof — here's how it works and why.

## More
- [GitHub (warden-protocol)](https://github.com/warden-protocol): Open-source repositories.
- [Warden Protocol](https://wardenprotocol.org): The team behind Halo.