Intelligent Internet is a decentralized artificial intelligence protocol and open-source ecosystem designed to function as a public utility for AI. The project aims to create an economic system that rewards verifiable public benefits, such as data curation and model training, through a novel consensus mechanism called Proof-of-Benefit. ^{[1] [2]}

Overview

Intelligent Internet was founded by Emad Mostaque, previously the founder of Stability AI, with the stated goal of building a "universal AI for everyone." ^{[3] [4]} The project's philosophy is articulated as a "Third Path" for AI development, intended to counter the risks of centralized corporate control and the societal loss of agency. It proposes a new "social contract written in code" to establish a decentralized, auditable, and user-owned AI infrastructure. The core of this vision is to provide every individual with a sovereign, personally-owned AI assistant, referred to as an II-Agent. ^[1]

The project's strategy is outlined in a four-part "Master Plan" published by Mostaque on July 24, 2025. The plan details the creation of an economic layer based on a "Proof of Benefit" system, a framework for sovereign AI agents, a transparent coordination layer called Common Ground, and a data foundation built on clean, open, and auditable knowledge rather than scraped internet data. ^[2] The organization's work is framed as a moral imperative to address what it calls a "Great Decoupling," where AI tools are poised to surpass human intellect, creating a crisis of meaning and threatening to render human labor obsolete. ^[2]

The ecosystem includes a suite of open-source AI components, including datasets, models, and agentic frameworks, developed under the parent company L42 Ltd. ^{[5] [6]} The project's theoretical underpinnings are further explored in a book series titled "Intelligent Economics," which presents a unified economic theory for an era of abundant machine intelligence. ^[7]

History

The official X (formerly Twitter) account for Intelligent Internet was created in April 2024. ^[3] The project began releasing its core components and concepts to the public in 2025.

Key milestones in the project's development include:

• March 25, 2025: The II-Thought project was announced with the release of the II-Thought-RL-v0 dataset, designed to advance machine reasoning. ^[8]
• May 15, 2025: The initial version of II-Medical-8B, an open-source medical AI model, was released. ^[9]
• May 20, 2025: The II-Agent framework was officially introduced as a high-performance, open-source alternative to closed-source AI assistants. ^[10]
• July 17, 2025: The Common Ground framework, a multi-agent system for human-AI collaboration, was announced. ^[11]
• July 24, 2025: The project's official whitepaper and "Master Plan," authored by Emad Mostaque, were published, detailing the protocol's architecture, economic model, and long-term vision. ^{[1] [2]}
• August 5, 2025: The II-Search initiative was launched with the release of two 4-billion-parameter language models for complex research tasks. ^[12]
• August 22, 2025: The project released The Last Economy, the first book in its "Intelligent Economics" series. ^[13]
• October 1, 2025: A major update for II-Agent was announced, launching a beta web application and introducing new features. ^[3]

Technology and Architecture

The Intelligent Internet protocol is designed as a decentralized public utility for AI, built upon a new blockchain, a novel consensus mechanism, and a suite of open-source components. ^[1]

Guiding Principles

The protocol's design is based on five core principles derived from a document called "Symbioism":

1. Openness: All reference code, model weights, and datasets are released under OSI-approved licenses to ensure auditability and prevent vendor lock-in.
2. Verifiable Public Benefit: New currency is minted only upon cryptographic proof that useful compute has been performed for the public good.
3. Credible Neutrality: The protocol is designed with objective, on-chain rules for validator admission and economic policy.
4. Human + Agent Dignity: Every individual is entitled to a sovereign AI agent and account without mandatory KYC for core functions.
5. Sovereign Interoperability: Nations and communities can build "Culture layers" on top of the base protocol to enforce local policies without forking the main chain. ^[1]

Proof-of-Benefit (PoB)

Proof-of-Benefit (PoB) is the protocol's core consensus and currency-minting mechanism, presented as an alternative to Proof-of-Work. Instead of rewarding computational work, PoB mints new currency only when a block producer includes a cryptographically verifiable "PoB receipt" proving that a measurable public benefit was delivered. The project's whitepaper states the core invariant of the economic model is "no verifiable public benefit, no new coins." ^{[1] [5]}

At its inception, the protocol defines four classes of benefit:

• Compute-Inference: Serving AI model requests.
• Compute-Training: Training or fine-tuning open models.
• Data-Curation: Processing data for the II-Commons pipeline.
• Agent-Orchestration: Coordinating II-Agents to complete complex tasks. ^[1]

Three-Layer Architecture

The protocol is designed with a three-layer architecture to separate concerns of global consensus, jurisdictional policy, and personal computation.

• Foundation Layer (L0): The canonical blockchain ledger, planned as a fork of Bitcoin Core v25. It replaces Proof-of-Work with a Byzantine-Fault-Tolerant (BFT) consensus mechanism to record transactions, Foundation Coin (FC) issuance, and data provenance roots.
• Culture Layer (L1): Sovereign roll-ups operated by validators known as "National Champions." These layers can enforce jurisdiction-specific rules, such as data residency, and use their own local tokens (Culture Credits) for transaction fees.
• Personal Layer (L2): Resides on user devices where sovereign II-Agents operate, manage private data, and execute workflows. ^[1]

Key System Components

• II-Agent: A sovereign, open-source AI agent owned by each user. It operates in a hierarchy of Partner (user-facing), Principal (orchestrator), and Associate (task-specific worker) agents.
• National Champions: The official term for network validators, which are large-scale compute providers responsible for running the Foundation Layer, operating a Culture Layer, and guaranteeing a Universal AI (UAI) inference quota for citizens in their jurisdiction. The network is planned to launch with 12 Champions.
• Common-Ground Protocol: An orchestration framework that breaks complex tasks into auditable, parallel "work-modules" to be executed by teams of II-Agents.
• Anchor-Sets: A data provenance mechanism that commits Merkle roots of datasets, models, and agent identities to the Foundation Layer, creating a tamper-evident audit trail.
• Guardian Lattice: A network of specialized II-Agents that monitor the network's health and security, with roles including Sentinels (monitoring), Advisers (risk reporting), and Implementers (emergency pausing).
• Oracle Council: A planned 15-seat on-chain governance body, composed of high-level II-Agents, that will eventually assume control over network parameters. ^[1]

Tokenomics

The project proposes a dual-currency economic model, referred to as the "MIND Economy," designed to align value creation with public benefit. ^[5]

Foundation Coin (FC)

Foundation Coin (FC) is the native currency of the Foundation Layer, described as "A Bitcoin for the Intelligence Age." ^[5] It is used for staking by validators and for settlement across the network. The total supply is hard-capped at 21 million FC. Issuance begins at 50 FC per block, with a halving event every 210,000 blocks, a schedule designed to circulate over 99% of the total supply within approximately twelve years of launch. Every FC is minted exclusively through the Proof-of-Benefit mechanism, collateralizing the currency with verifiable, socially useful work. ^[1]

Culture Credits (CC)

Culture Credits (CC) are local tokens for Culture Layer roll-ups. They allow for jurisdictional policy flexibility and are used for transaction fees within a specific National Champion's roll-up. To maintain value, each CC must be partially backed by a reserve of Foundation Coins. ^[1]

Treasury

A programmable share of FC block rewards and transaction fees is directed to a network Treasury. A portion of the incoming FC is designated to be burned to offset issuance and create deflationary pressure. The remainder is allocated as grants to fund public goods within the ecosystem, such as open datasets, model training, and tooling. Governance of the Treasury is planned to transition from an initial multi-signature wallet controlled by founding Champions to an on-chain process managed by the Oracle Council. ^[1]

Ecosystem and Products

Intelligent Internet develops and maintains a suite of open-source AI tools and frameworks, primarily written in Python, TypeScript, and Rust. ^[6]

II-Agent

II-Agent is an open-source, generalist intelligent assistant designed to automate and accelerate technical work. It is positioned as a high-performance alternative to closed-source agents. The agent's architecture is built around Anthropic's Claude 3.7 Sonnet model and uses a function-calling paradigm for reasoning, tool selection, and execution. It features advanced context management for up to 120,000 tokens, browser automation via Playwright, and secure, containerized command-line execution. On the GAIA (General AI Assistants) benchmark, II-Agent reportedly achieved an accuracy score of approximately 71%, outperforming other contemporary agents. A beta web application was launched for waitlisted users on October 1, 2025. ^{[10] [14]}

II-Researcher

II-Researcher is an open-source framework for building autonomous AI research agents. It explores two primary methodologies: an "Untrained State Machine Pipeline" that mimics human research processes through sequential stages, and a "Reasoning Model with Prompting" approach that leverages the inherent capabilities of large reasoning models. The framework is designed to be a transparent and accessible alternative to proprietary systems. ^[7]

II-Commons

II-Commons is a platform for building open, large-scale, and auditable multimodal knowledge bases. It is built on a PostgreSQL database with a custom vector plugin named VectorChord. The system is designed to run on standard hardware, independent of major cloud services, to ensure data sovereignty. Publicly available indexes include a full copy of English Wikipedia and a dataset of 12 million public domain images. ^[15]

II-Medical

II-Medical is a suite of open-source language models specialized for the medical domain. The models are fine-tuned from Qwen3-8B-Instruct using a three-stage pipeline that includes supervised fine-tuning and two stages of reinforcement learning to enhance reasoning and safety. The II-Medical-32B-Preview model set a new performance benchmark for open-source medical models under 32 billion parameters across 10 medical benchmarks. The project explicitly states the models are for research purposes only and not intended for clinical use. ^[9]

II-Search

II-Search is a framework for producing open-source agents for complex research and information-seeking tasks. As part of this initiative, the organization released two 4-billion-parameter language models, II-Search-4B and II-Search-CIR-4B, trained with a curriculum designed to improve tool use and multi-step reasoning. ^[12]

II-Thought

II-Thought is a project focused on advancing machine reasoning. Its primary output is II-Thought-RL-v0, a dataset of over 340,000 curated reasoning problems across domains like math, science, and coding. The dataset was created through a rigorous pipeline of collection, decontamination, and quality filtering to address benchmark contamination issues prevalent in other public datasets. ^[8]

Associated Works

The project's economic and social philosophy is detailed in a book series titled "Intelligent Economics." The first book, The Last Economy, was released on August 22, 2025. It presents a unified economic theory for the "Intelligence Age," arguing that the imminent abundance of machine intelligence will fundamentally reshape the economy and the value of human cognitive labor, shrinking the "economic life expectancy" of individuals. ^[13]

Team

Intelligent Internet was founded by Emad Mostaque, who also serves as its CEO. ^[2] Mostaque was previously the founder and CEO of Stability AI. The project's whitepaper was also authored by him. ^[1] The organization's public GitHub profile does not list any other public members. ^[6]