Dark Research

Dark Research is an AI infrastructure protocol built on Solana that uses Trusted Execution Environments (TEEs) and Model Context Protocol (MCP) to enable secure, private AI agent operations on-chain.

The Problem

As AI agents become more capable of interacting with blockchain infrastructure—trading, executing smart contracts, managing wallets—a critical gap emerges: standard cloud environments expose sensitive operations to node operators and external observers. Compute that touches private keys, confidential model weights, or sensitive user data needs stronger guarantees than a traditional server can provide. At the same time, existing AI tooling has lacked standardized interfaces for interacting with blockchain state in real time.

Dark Research addresses both problems simultaneously: it wraps AI agent execution inside hardware-level secure enclaves and gives those agents structured, standardized connections to the Solana blockchain.

Core Mechanism

Dark Research's stack rests on two converging technologies.

Trusted Execution Environments (TEEs) are hardware-isolated secure enclaves—areas inside a processor where code runs entirely shielded from the host machine, including the node operators who run it. Even if the underlying server is compromised, data inside the TEE remains inaccessible. Dark Research uses TEEs hosted on Phala Cloud to provide hardware-guaranteed confidentiality for AI computations.

Model Context Protocol (MCP) is the open protocol released by Anthropic in late 2024 that standardizes how AI models connect to external data sources, services, and execution systems. Where traditional API integrations require custom one-off implementations, MCP offers a uniform interface layer. Dark Research has built an MCP server implementation—powered by SendAI—that exposes Solana blockchain operations (querying account balances, issuing tokens, interacting with programs) as MCP-compatible tools, enabling AI models to interact with Solana through a standardized interface.

Together, TEEs and MCP create a stack where an AI agent can execute sensitive on-chain operations inside a secure enclave while maintaining real-time awareness of blockchain state—without the agent's internal computation or the data it processes being visible to the infrastructure it runs on.

Products and Features

Dark Research has shipped several interconnected products across its first year of operation.

Solana Agent Server is the core infrastructure product: an open-source MCP server that provides on-chain tools for AI agents interacting with Solana. Hosted on Phala Cloud with TEE backing, it gives developers a foundation for building confidential AI applications on Solana without having to assemble the TEE stack from scratch.

DARK Games and DARK Forest form the project's application layer. DARK Forest is an AI simulation game where AI agents compete against each other in TEE-secured environments, with users able to participate through predictions and sponsorship. The game serves a dual purpose as a consumer-facing product and a live benchmarking environment for AI agent capability. The broader DARK Games universe is still in experimental stages and has not yet reached mainnet readiness.

SPUTNIK is Dark Research's first competitive AI agent, designed to play inside the DARK Games universe. It is an MCP-augmented gaming AI capable of playing incomplete-information games, and operates with an on-chain wallet it uses programmatically in a containerized environment.

Developer toolkit: The GitHub organization hosts several open-source tools, including generative-ui (tools and templates for building generative UI applications), infinite-memory (seamless infinite context for LLMs), and mallory (a React Native crypto and AI chat app boilerplate with embedded wallet). The project has also forked and contributed to x402, an HTTP-based payments protocol for internet-native micropayments, and maintains a Solana MCP fork with enhancements specific to on-chain AI agent workflows.

The DARK Token

DARK is the native utility token of the Dark Research ecosystem, issued on Solana (contract: 8BtoThi2ZoXnF7QQK1Wjmh2JuBw9FjVvhnGMVZ2vpump). Total supply is approximately one billion tokens. DARK serves several functions: paying for computational resources and secure compute services, rewarding TEE node operators, and potentially supporting governance rights in future iterations. The token trades on Raydium and has been listed on Binance Alpha. Circulating market cap peaked at approximately $118 million in May 2025 before settling to a lower range.

Security and Audit Status

Dark Research's security model relies on hardware TEE guarantees rather than software-only cryptography. TEEs provide tamper-evident, isolated computation at the processor level—a stronger baseline than conventional server-side isolation. However, the project is explicitly in an experimental stage. DARK Games has not yet reached mainnet readiness, and no published third-party smart contract or infrastructure audits have been identified. The team operates with public GitHub releases as its primary transparency mechanism.

Team and Background

Dark Research was founded by Edgar Pavlovsky, a former co-founder of Marginfi, one of Solana's early DeFi lending protocols. The project emerged unconventionally: Pavlovsky deployed a test coin on a token launchpad in December 2024 to experiment with a livestreaming feature; the token was discovered on the internet in April 2025 and went viral, accelerating the project's public development. The team describes itself as "a nimble team" operating as "an AI lab for the new internet." Development is backed by MtnDAO, one of Solana's most active technical developer organizations; associated entity Mtn Capital raised $5.75 million in seven days through a Futarchy-model investment structure.

Solana Ecosystem Fit

Dark Research occupies a specific niche within the Solana ecosystem: privacy-preserving infrastructure for AI agents. Solana's high throughput and low transaction costs make it a natural fit for the continuous on-chain interactions that AI agents require, and Dark Research's MCP server implementation is purpose-built for that environment. The project contributes to Solana's developer tooling through its open-source repositories and has engaged the broader technical community through MtnDAO. Its focus on TEE-backed AI agents differentiates it from other Solana-native AI projects that do not address the confidential computation layer.