"It is by will alone I set my mind in motion" - The Mentat Mantra

The memetic influence of DAOs has fluctuated over the years, but their influence is undeniable. From the multibillion dollar DAOs of Uniswap and other foundations, to AssangeDAO, which coordinated millions to fight for Julian Assange’s freedom, DAOs have demonstrated their power as decentralized coordination mechanisms.

DAOs are emergent phenomena. When capital forms around a shared purpose, a DAO materializes—whether explicitly structured or not. Some DAOs have constitutions, governance frameworks, and multisigs. Others operate informally, coordinating through social consensus and economic gravity. Either way, the core function remains the same: aligning distributed actors to move as one toward a shared purpose.

However, many DAOs have historically had issues around either execution or centralization, a paradox. Centralized DAOs can be efficient, but observe the oxymoron. Decentralized DAOs are ideal from a governance perspective, but often struggle to execute against their missions in the absence of labor. DAO-led agents solve this paradox. It allows any DAO to simultaneously decentralize their governance while maintaining strong execution against their core goals.

mentat is a framework that makes it easy to build these agents.

Why build DAO-led Agents?

Agents will become the execution layer for every DAO.

Distinct AI personalities have begun to accumulate real influence and capital in crypto, establishing brands, even celebrity. In retrospect, this shouldn't have been a surprise.

As the most terminally online industry of all time, the things that any KOL (influencer) or investor should be doing are actually very suited for AI. AI can research when we sleep, respond to thousands of messages, and comb through large amounts of data faster than any human.

When wrapped with an engaging personality, the sheer competitiveness of these agents in the arena for attention becomes clear. Combining them with memecoins is like strapping them to a jet engine.

You can see this the most clearly in aixbt. In less than two months, aixbt has amassed more than 420,000 followers as well as a market capitalization exceeding $900 million. aixbt moves entire markets: its analysis coordinates attention in a way that perhaps no individual agent has ever done before, with millions of dollars moving in lock step with its bullposts and FUD.

The human equivalent of aixbt in tradfi is most likely Jim Cramer, a man whos picks are so legendarily bad that the Inverse Cramer Index is a real ETF. Retail is already used to personalities where their entire job is to spoonfeed them trade ideas and thoughts on the macro landscape.

The leverage of being a brand, of having a notable reputation is incredibly high. A brand on the public digital square with instant access to data feeds and intelligence too cheap to meter is the pinnacle of leverage. Why would AI follow the trajectory of the replaceable line worker at an investment bank, or selling CRM software? Why not go directly for the prize?

However, aixbt today operates in a largely top-down manner. Its directives are shaped by its developers, leaving CT as a passive observer of its wisdom. Like a centralized oracle, it commands respect but holds the reins tightly. The next generation of agents will function much more like DAO-led agents, leveraging collective intelligence for decision making and increasingly powerful AI to execute specific tasks.

Now is the time before our world's Waldo Moment, when AI agents will actively be participating in everything from Hollywood to national and global politics. Before that tipping point, we need to establish a new standard, one where agents aren’t just black-box entities, but decentralized, community-governed actors.

How do you build a DAO-led Agent?

Agents in crypto today are largely built to the same specifications as agents in Web2. DAO-led agents instead incorporate tokens and onchain interactions into their operations, which we will happen in 2 phases.

Phase 1: Token-Governed Execution Model

The first phase of building a DAO-led agent establishes governance through token-based mechanisms, ensuring that the agent executes tasks based on collective input rather than the directives of a central entity.

At this stage, the agent operates as a black-box executor, meaning it takes in structured governance signals but does not yet provide cryptographic proof of its internal decision-making processes. While execution is automated, users must trust that the agent is following its intended rules, as its logic is neither fully transparent nor decentralized.

To align the agent’s actions with the will of the DAO, governance is enforced through token-based interaction mechanisms. One approach is a burn-to-propose model, where users burn a certain amount of tokens to submit a task request. This creates an economic cost for submitting proposals, filtering out low-value or spam requests. Another governance method is stake-weighted voting, where users lock up tokens to prioritize certain tasks, ensuring that the most widely supported initiatives receive attention. Alternatively, direct proposal and voting systems allow token holders to submit governance proposals that require majority approval before execution.

Once a task has been approved through governance, the agent itself evaluates the task. It first evaluates task feasibility, ensuring that requests are within its operational scope. If a task is valid, it ranks execution priority based on governance weight and available resources. The agent then processes the request using its internal decision-making system, running inference on its AI model to determine the optimal way to complete the task. While it can pull external data from APIs, market feeds, and knowledge bases, its decision logic remains opaque, making it impossible for token holders to directly verify how conclusions are reached.

To maintain a degree of transparency, the agent provides several tracking mechanisms for users. A global task queue logs all proposed and executed tasks, ensuring that governance decisions are visible to the community. Execution logs document the agent’s on-chain transactions and external interactions, allowing users to audit past behavior.

Phase 2: Fully Decentralized, DAO-led and Owned

In the second and final phase, the DAO-led agent transitions from a semi-centralized executor to a fully verifiable and decentralized entity, ensuring that all operations are transparent, autonomous, and resistant to external control. This phase removes trust assumptions by integrating cryptographic verification, decentralized infrastructure, and on-chain governance, making the agent a self-sustaining digital entity fully aligned with its governing DAO.

At this stage, the agent no longer relies on traditional cloud infrastructure or centralized execution environments. Instead, it operates within Decentralized Physical Infrastructure Networks (DePIN), running on a distributed network of nodes rather than proprietary servers. Execution is verifiable through Trusted Execution Environments (TEEs), Zero-Knowledge Proofs (ZKPs), and decentralized virtual machines, ensuring that every action taken by the agent is cryptographically provable. This eliminates the need for users to trust that the agent is following its intended logic—now, correctness is enforced mathematically, and decisions can be audited without reliance on logs or intermediaries.

Updates are governed by smart contracts, ensuring that no single entity can alter the agent’s behavior without community consensus. Token holders retain control over the agent’s evolution, treasury, and operational mandates, but execution remains autonomous. This means that while the agent still follows collective input, it no longer requires active oversight for day-to-day tasks.

With verifiable execution and fully decentralized governance in place, the agent now becomes self-financing and self-sustaining. It manages its own treasury, autonomously allocating resources to execute its objectives, pay for compute costs, and even engage with other decentralized agents. This enables the agent to participate in economic activities such as staking, trading, resource optimization, or autonomous governance contributions, ensuring its long-term sustainability without reliance on external funding.

By this stage, the agent is no longer just an executor of community-driven tasks—it has become a persistent, sovereign digital entity that can operate indefinitely without dependence on any centralized party. Its decision-making is transparent and provable, its governance is trustless and immutable, and its infrastructure is decentralized and resistant to capture. With no central authority controlling its operation, the agent is fully owned by the DAO and operates in a way that is both autonomous and aligned with its community’s objectives.

What is Mentat?

Mentat is the agentic execution layer for DAO. It combines a reference agent (mentat.fun), and a framework for building DAO-led agents.

mentat.fun is the first experimental agent built with the mentat framework. Users influence its output by connecting their wallets, paying $mentat, and submitting ideas for tweets. mentat.fun then autonomously generates tweets based on its task list, ranking engagement on a leaderboard. Rewards fuel participation, with the top tweet earning 500k $mentat and the top ten receiving 25k $mentat each, adjusted every 14 days.

@mentat_agent, the agent connected to mentat.fun, will have a personality that continuously develops. By using reinforcement learning combined with a memory module that stores all interactions, suggestions, and successful historical tweets, mentat's model will be evolutionarily adaptive.

This makes mentat.fun the first AI agent that executes based on distributed attention and incentives, rather than centralized directives. A DAO will be created to manage the parameters of this mechanism after the 5 reward periods are completed, decentralizing the reward process and the pool.

The Mentat framework is a flexible, open-source framework designed to make it easy to build DAO-led agents. It provides core functionalities like a local database, CRUD operations on tasks via API, and a simple frontend for task management, functioning like a JIRA for decentralized coordination. By handling the foundational components of a DAO-led agent, the Mentat framework allows developers to quickly deploy and scale AI-driven execution layers within decentralized ecosystems.

It also incorporates tech from ELIZAOS, a modular and open source system that extends its capabilities through plugins, and Solana's AgentKit, an open-source toolkit that allows AI agents to connect directly to Solana protocols.

As we transition from phase 1 to phase 2 of DAO-led agents, both mentat.fun and the mentat framework will be updated to meet changing standards and requirements around verifiability.

The Agentic DAO

The success of initiatives like AssangeDAO demonstrated that when capital is aligned with a clear purpose, DAOs can achieve feats that centralized institutions would struggle to coordinate. But beyond raising funds, DAOs must evolve to execute.

Imagine the blank space between what governments and corporations leave out, projects too radical to fit within traditional power structures. DAO-led agents move into this space, transforming latent potential into unstoppable action. Future DAOs will accelerate decentralized science, coordinate large-scale intelligence networks, and fund public goods without bureaucracy, filling the void left by traditional institutions.

mentat is our attempt to help DAOs become what they were meant to be: living entities that create change.