Eliza Framework Explained: AI16Z’s Rise, Fall and Lawsuit

Type a personality into a JSON file, point it at an X account and a Solana wallet, and within minutes you can have a bot that posts, replies and trades on its own. That is the basic pitch behind Eliza, the open-source framework that turned a roughly $75,000 Discord experiment into software running thousands of AI agents, and eventually into a federal class action lawsuit. Eighteen months separate those two facts, and the distance between them says as much about the limits of “autonomous” AI in crypto as it does about the technology itself.

This is a look at what the Eliza framework actually does, how the ai16z project built on top of it became one of the defining (and now most litigated) stories of the AI-agent boom, and where the technology stands separately from the drama around its best-known application. The story matters beyond one token’s chart, because a meaningful share of the AI-agent sector is built, directly or indirectly, on the same toolkit.

What Is the Eliza Framework?

Eliza is an open-source, TypeScript framework for building AI agents that keep persistent memory, connect to social platforms such as Discord, Telegram and X, and act on blockchains including Solana, Ethereum, Base and BNB Chain. It is maintained by Eliza Labs and the elizaOS organization on GitHub, where the core repository has drawn more than 18,700 stars and 5,600 forks, putting it among the most-starred AI agent repositories in open source, according to the project’s GitHub page. That contributor base is unusually broad for a two-year-old project: pull requests have come from independent developers building plugins for chains and services the core team never touched directly, generally a sign of a framework that has outgrown its original use case rather than one being kept alive artificially by its own creators.

In practice, that toolkit gets used for a wide range of things that have nothing to do with trading. Community moderators that answer support questions in a project’s Discord, research agents that summarize governance proposals before a DAO vote, and content bots that draft social posts from a fixed personality are all common Eliza deployments, largely because the character-file and plugin system make it fast to stand up a narrow, well-scoped bot rather than a general-purpose one. The trading and treasury-management use case that made ai16z famous is a small, high-risk slice of what the framework is actually used for day to day.

The framework’s runtime is model-agnostic (it can call OpenAI, Anthropic, Gemini, Grok, Llama or a self-hosted model) and built around four primitives: actions define what an agent can do, such as replying to a message or executing a token swap; providers inject context into the agent’s prompt, like a wallet balance or recent chat history; evaluators run after a response to update memory or flag content for review; and services handle long-running connections, such as a database or a Discord client. An agent’s personality, backstory and example dialogue live in a “character file,” a JSON document the runtime loads at startup.

That architecture is documented in the project’s own technical paper, published in January 2025, which frames Eliza as Web3-native infrastructure rather than a general chatbot toolkit. It began, however, as the backbone of a much narrower and stranger experiment: an AI agent built to imitate, and poke fun at, one of Silicon Valley’s most famous venture capital firms.

From a Discord Side Project to a Multi-Billion-Dollar Token

On October 24, 2024, developer Shaw Walters launched ai16z on daos.fun, a Solana-based platform for crowdfunded investment DAOs. The idea was blunt: build an AI facsimile of Andreessen Horowitz, name it after firm co-founder Marc Andreessen, and let a bot narrate crypto investment calls informed by token-holder suggestions. The initial funding target was small, around $75,000.

Three days later, the joke reached its target. On October 27, 2024, Marc Andreessen himself posted “GAUNTLET THROWN” on X alongside a screenshot of the bot’s profile. He never invested in or endorsed the project, but the shoutout alone pushed ai16z toward an early market cap of roughly $96.6 million within hours, according to The Block’s reporting at the time.

Three days after that, the story got more complicated. On October 30, 2024, crypto outlet Protos published an investigation arguing the project’s “autonomous AI” breakthrough was, in practice, human-operated: a person was writing and approving the bot’s calls rather than a model executing them independently, a pattern the outlet described as a form of AI-washing dressing conventional operations in autonomous-agent branding, according to Protos. The gap between the marketing and the mechanics would resurface eighteen months later as the centerpiece of a class-action complaint.

None of that slowed the token in the short term. Through the AI-agent mania of that winter, the price climbed until it peaked at $2.47 on January 2, 2025, giving the token a fully diluted valuation above $2.7 billion, according to historical data on CoinGecko. For a few weeks, a bot built to parody a venture firm was, on paper, worth more than most of the startups that firm had funded.

Inside the Architecture: Actions, Providers and Plugins

Strip away the token and the drama, and Eliza is a fairly conventional agent framework wrapped around blockchain-specific tooling. The runtime loads a character file, matches an incoming trigger (a Discord message, an X mention, a scheduled task) to a registered action, pulls supporting context from providers, executes the action, and then runs evaluators that can update memory or flag a response before it goes out. This structure is laid out in the project’s technical paper, “Eliza: A Web3 friendly AI Agent Operating System,” published on arXiv in January 2025 by Walters and thirteen co-authors from Eliza Labs.

• Actions: what an agent can do, from posting a reply to executing an on-chain swap
• Providers: context fed into the agent’s prompt, such as wallet balances or recent messages
• Evaluators: post-response logic that updates memory or reviews output
• Services: long-running connections, like a database adapter or a Telegram client

The plugin registry lists more than 90 official integrations spanning Discord, Telegram, X, Ethereum, Solana and OpenAI-compatible model providers, according to elizaOS’s own documentation. Developers can run an agent locally, in Docker, or on a hosted Eliza Cloud service, and swap plugins through a package-registry CLI without touching core code.

What the framework does not do is give an agent better judgment than the language model wired into it. Eliza gives that model hands (wallets, APIs, social accounts) and a memory. Whether those hands should be trusted with real money is a separate question, and it is one the framework’s own creator has answered, in public, more cautiously than the marketing around his own flagship project implied he would.

How Autonomous Are These Agents, Really?

Walters has said in interviews that an Eliza-based agent’s near-term value lies less in making fully independent investment calls and more in structuring market data, reading social sentiment and executing trades faster than a person could, and that projects built on the framework typically run inside smart-contract guardrails and preset limits rather than being handed unrestricted control of funds. He has framed the technology’s near-term role in finance and DAOs as augmenting human coordination rather than replacing it outright.

That framing sits uneasily next to how ai16z was marketed at launch, as a fund where an AI, not a person, decided what to buy. The Protos investigation, and later the class-action complaint built partly on it, argued the gap between the pitch and the mechanics was not incidental: that dressing a conventionally operated fund in autonomous-agent language was what let a five-figure raise become a token worth billions on paper within about ten weeks.

The ai16z to ElizaOS Rebrand

By January 2025, the joke had stopped being funny to the firm it referenced. Chris Dixon, managing partner at a16z crypto, said on the Unchained podcast that Andreessen Horowitz had asked the team to change the project’s name. “We were not connected with AI16z,” Dixon said. “I believe they are in the process of changing it. We’ve asked them to. It just was creating a little bit of confusion,” according to Unchained’s reporting. Dixon added that the firm still welcomed the broader experiment: “I love to see the experimentation with all the AI agent bots. This is what’s fun about permissionless innovation.”

Walters announced the rebrand to ElizaOS in the project’s Discord that Sunday evening. The token ticker itself stayed AI16Z for the time being, with any change left to a future DAO vote once a governance module launched. That vote, and the actual token migration, would not arrive until the following November, by which point the price had fallen a long way from its January peak.

ElizaOS v2 and the Plugin Ecosystem

The ElizaOS v2 beta shipped on March 18, 2025, rebuilding the runtime around an event-driven architecture and adding hierarchical task decomposition, letting an agent break a broad goal, such as growing a community or managing a treasury, into smaller ordered subtasks it can adjust as conditions change. It replaced a more tangled v1 codebase that had grown quickly and somewhat organically during the 2024 hype cycle, and it introduced a cleaner package-registry system so plugins could be added or swapped without editing core runtime code.

Around that same architecture, Eliza Labs built The Org, a reference implementation showing several specialized agents, a community moderator, a developer-relations bot, a project coordinator and a social media manager, working inside one Discord server. It is an early, fairly modest sketch of the “autonomous company” pitch that shows up across the AI-agent sector, closer to a proof of concept than a finished product.

Chainlink CCIP and the Cross-Chain Push

On May 19, 2025, Chainlink’s Cross-Chain Interoperability Protocol went live on Solana, and elizaOS was named among the first frameworks adopting the associated Cross-Chain Token standard, according to the official Chainlink press release. “CCIP and the CCT Standard allow us to extend agent capabilities and assets across leading blockchain ecosystems, while continuing to leverage Solana’s high performance and efficiency,” said Shaw Walters, founder of elizaOS, in the announcement. “This integration brings us closer to a future where AI agents can transact, coordinate, and deliver value across chains leveraging trust-minimized infrastructure.”

In practice, that means an Eliza-based agent can hold and move assets across Solana, Ethereum, Base and BNB Chain without relying on a conventional token bridge, historically one of the most exploited categories of smart contract in crypto. Bridges of that kind have produced some of the largest single losses in crypto history, since they typically concentrate enormous value behind a small set of validating keys or contracts. For a framework whose whole premise is letting semi-autonomous software move real value, cutting that dependency out of the equation is not a cosmetic feature, even though it does nothing to protect the private key controlling the agent’s own wallet, which remains the more immediate point of failure.

The Token Migration That Became a Lawsuit

Between October and November 2025, the team finally executed the token migration the January 2025 rebrand had deferred, swapping AI16Z for a new ELIZAOS token. According to the class-action complaint later filed against the project, total supply expanded roughly tenfold, from about 1.1 billion to 11 billion tokens, but only 60% of the new supply went to existing AI16Z holders converting their positions. The remaining 40%, the complaint alleges, was split between undisclosed private investors (15%), team members and contributors (10%) and entities the complaint says the defendants controlled, diluting existing holders without prior public disclosure of the split, according to a summary of the filing published by ClaimDepot.

By the time of the migration, the price had already fallen enormously from its January 2025 peak. The new token gave the project a way to reset its own chart and branding, but it also gave plaintiffs’ lawyers a specific, dated transaction to point to as evidence of dilution.

Inside the Lawsuit: What Burwick Law Is Alleging

On April 20 to 21, 2026, plaintiff Gorka Pikabea, a resident of Spain, filed a federal class action in the U.S. District Court for the Southern District of New York, case number 1:26-cv-03238, naming Eliza Labs Inc., Shaw Walters, company secretary Sebastian Quinn-Watson and the AI16Z DAO as defendants, alongside fifty unnamed Doe defendants, according to court filings summarized by ClaimDepot.

The complaint centers on naming as much as on financial mechanics. It alleges the project chose “AI16Z” specifically to trade on Andreessen Horowitz’s reputation and named its flagship bot “Marc AIndreessen” without authorization from the firm or from Andreessen himself. It leans heavily on the October 2024 Protos reporting to argue the autonomous-agent framing was misleading from day one, and it did not stop there: the class covers everyone who bought AI16Z and migrated to ELIZAOS between October 24, 2024 and April 20, 2026, a group the complaint says includes at least 3,945 wallet addresses showing net losses.

The allegations track a pattern that investors and journalists have flagged across the wider AI-agent sector: a project markets a bot as self-directed to generate excitement and token demand, while people quietly handle the parts that actually require judgment. Whether that pattern amounts to actionable fraud, false advertising, or simply aggressive but lawful marketing is exactly what a court is now being asked to sort out, and the ai16z case is among the first to reach a filed complaint rather than just online criticism and outlet investigations.

• Two claims under New York’s General Business Law (deceptive business practices and false advertising)
• Two claims under California’s Unfair Competition Law and False Advertising Law
• Additional common-law claims including unjust enrichment
• Relief sought: actual and treble damages, restitution, disgorgement of profits, an injunction and attorneys’ fees

Notably, the suit is built on state consumer-protection and false-advertising statutes rather than federal securities claims. None of the defendants had filed a public response as of publication, and the allegations remain unproven in court.

The X Corp dispute shows Eliza Labs is willing to litigate on its own behalf, too. In August 2025, the company sued Elon Musk’s X Corp in federal court in San Francisco, alleging X induced it to share confidential technical details under the guise of a partnership, then suspended Eliza’s accounts and demanded an enterprise license costing $50,000 a month before it would restore access, according to The Block’s coverage of the filing. That case was later dismissed with prejudice after the two sides reached a private resolution; according to reporting on the dismissal, neither X nor its AI unit xAI paid Eliza Labs anything, and Eliza instead purchased a standard enterprise API license.

What the Numbers Look Like Now

The scale of the collapse is easiest to see in plain figures. As of this writing, ELIZAOS trades for a fraction of a cent, with a market capitalization of roughly $4 million, and the token set a fresh all-time low just days before publication, according to CoinGecko. Figures vary slightly across data aggregators depending on which supply figure they use, but the broad picture is consistent: a token that briefly carried a fully diluted valuation above $2.7 billion in January 2025 is now worth roughly a thousandth of one percent of that peak.

ELIZAOS trades on a handful of centralized and decentralized venues rather than every major exchange. For readers deciding where to actually buy, sell or custody a small-cap agent token like this one, HOGE Wire’s guide to judging any exchange is a reasonable checklist to run through before sending funds anywhere.

Eliza Framework vs Virtuals Protocol vs LangChain

Eliza is not the only agent framework crypto has produced, and its core problems have arguably been less about the technology than about what got built on top of it and how that project was marketed and financed. Three points of comparison are useful: Virtuals Protocol, a competing crypto-native launchpad; and LangChain, the dominant general-purpose framework for LLM applications outside crypto entirely.

LangChain dwarfs both crypto frameworks by conventional open-source metrics, with more than 141,000 GitHub stars and over 23,500 forks behind its Python-based agent and LLM-orchestration tooling, according to its GitHub repository. It has no native token and no particular interest in blockchains, which is precisely why crypto-native teams still reach for Eliza or Virtuals instead: both ship wallet handling, on-chain actions and social-platform clients as first-class citizens rather than as something a developer bolts on.

Virtuals, for its part, has held up better as a token than ELIZAOS has, trading with a market capitalization in the mid hundreds of millions of dollars, according to CoinGecko, though it too sits far below its own highs from the 2024 to early 2025 AI-agent run. The structural difference is philosophical as much as technical: Virtuals bakes tokenization and revenue-sharing into how an agent gets created in the first place, while Eliza treats the token as something a project can bolt on top of an otherwise free framework, for better or worse.

There is also a governance difference worth flagging for anyone comparing the three as investments rather than as developer tools. LangChain is a conventional venture-backed company whose roadmap answers to investors and paying enterprise customers. Virtuals and Eliza both route at least some decisions through token-based structures: Virtuals through its own protocol governance, and Eliza, at least in its ai16z incarnation, through the AI16Z DAO now named in active litigation. A framework’s governance model says very little about its underlying code quality, but it says a great deal about who is accountable when something like the ELIZAOS token migration happens, which is precisely the question Burwick Law’s complaint is asking a court to answer.

Security: Wallets, Guardrails and the Custody Problem

Any Eliza-based agent that trades on-chain needs a private key, and that key has to live somewhere the runtime can reach it: a hot wallet, a multi-party-computation setup, or a smart-contract wallet with preset spending limits. That is a fundamentally different custody problem than a human trader reviewing every transaction before signing it, and it is the same tradeoff explored in HOGE Wire’s comparison of custody models: broadly, the more autonomy an agent has, the fewer chances exist for a human to catch a mistake before it becomes final and on-chain.

The custody options in practice span a spectrum. A hot wallet gives an agent unrestricted signing power and the fastest execution, which is also the riskiest setup if the underlying model is manipulated or the key itself leaks. A smart-contract wallet with programmed spending limits, allowlisted counterparties or daily caps trades away some flexibility for a hard ceiling on how much any single mistake can cost. Multi-party computation setups split the signing key across several parties so no single compromised process can move funds alone, at the cost of added latency and infrastructure. Walters’ own comments about keeping agents inside guardrails point toward the middle of that spectrum rather than the unrestricted hot-wallet extreme, even though ai16z’s early marketing implied something closer to full autonomy.

Security audits complicate the picture further. A third-party review of a plugin’s smart contracts can confirm the contract itself does what it claims, but it says nothing about whether the language model calling that contract will be tricked into calling it at the wrong time, with the wrong parameters, by a cleverly worded message. That distinction, between auditing code and auditing judgment, is exactly the gap HOGE Wire’s review of DeFi audit firms flags as an open problem across the industry, and it applies with extra force to any framework whose selling point is letting software make its own calls.

There also is not yet a mature, well-priced bug-bounty market for the specific failure mode of “the model was socially engineered,” the way there is for conventional smart-contract vulnerabilities. HOGE Wire’s look at how crypto bug bounties actually pay out shows a fairly established economy for finding a reentrancy bug or an oracle flaw; nothing comparable yet exists for finding the prompt that talks an agent into sending its wallet’s contents to a stranger.

Where Regulation Is Heading for Agent Frameworks

No regulator has a bespoke rulebook yet for software that holds a wallet and decides what to do with it. In the United States, the SEC’s posture has shifted noticeably since Paul Atkins became chair in 2025. Rather than treating autonomous trading tools as inherently suspect, Atkins has said the agency is weighing new rulemaking covering onchain trading systems, crypto vaults and other automated financial applications, framing AI agents as a coming feature of market structure rather than a threat to it, according to CoinDesk’s coverage of his May 2026 remarks.

That openness does not resolve the liability question the ai16z saga puts front and center: if an autonomous agent executes a bad trade, or is manipulated into sending funds to an attacker, who is actually responsible, the developer who wrote the code, the platform hosting the wallet, or the person who deployed the agent in the first place. For a broader look at how the agency actually builds and settles cases once it decides to act, HOGE Wire’s explainer on SEC crypto enforcement in 2026 is a useful companion piece.

The Burwick Law suit against Eliza Labs is itself a preview of where these accountability fights are actually being litigated in 2026: not through SEC securities claims, but through ordinary state consumer-protection and false-advertising statutes, the same category of law used against misleading product marketing generally, applied here to a piece of software marketed as having a mind of its own.

What Comes Next for Eliza

Despite the litigation and the collapsed chart, the underlying framework has not gone quiet. Development activity on GitHub has continued into 2026, and Eliza Labs has kept talking publicly about partnerships and enterprise interest that sit apart from the AI16Z DAO and its token, building on the credibility the project earned early on through its research partnership with Stanford’s Future of Digital Currency Initiative, announced in December 2024 and overseen by cryptographers Dan Boneh and David Mazieres, according to Blockworks.

ElizaOS has also found buyers outside crypto-native speculation. In 2026, a separate public company, Secure Blockchain, announced it had acquired an outside venture called Agentic Solutions and was using that acquisition to launch an enterprise agentic AI platform built on the open-source elizaOS framework, according to a company announcement distributed via GlobeNewswire, independent evidence that the underlying code has commercial value apart from the DAO that made it famous.

The clearest lesson from Eliza’s first eighteen months is that the framework and the token are two different things that got marketed as one. The open-source software genuinely works, is genuinely used by outside developers, and does not depend on any particular token’s price to keep functioning. The DAO and token built on top of it are now, separately, the subject of a lawsuit alleging the whole thing was less autonomous, and less honestly disclosed, than investors were led to believe. Both things can be true about the same project at once, and the next chapter of the AI-agent sector will likely turn on whether other teams can keep that distinction cleaner than ai16z did.

Frequently Asked Questions

What is the Eliza framework?

Eliza is an open-source, TypeScript framework built by Eliza Labs for creating AI agents that hold persistent memory, connect to platforms like Discord, Telegram and X, and interact with blockchains including Solana, Ethereum, Base and BNB Chain. It provides the plugins, character-file format and runtime that let a developer give a language model a name, a personality and a wallet without building that infrastructure from scratch. It is maintained by the elizaOS organization on GitHub and documented in a January 2025 technical paper on arXiv.

Is ElizaOS the same thing as ai16z?

They are related but not identical. Ai16z was the original Solana-based AI investment DAO and token launched in October 2024 that used the Eliza framework. ElizaOS is the rebranded name for both the open-source framework and the company and DAO behind it, adopted in January 2025 after Andreessen Horowitz asked the team to stop using a name that echoed its own. The AI16Z token itself did not actually migrate to the new ELIZAOS ticker until October and November 2025, and the two names are still used somewhat interchangeably across exchanges, data aggregators and older articles, which is a common source of confusion for anyone researching the project after the fact.

What happened to the ai16z and ELIZAOS token price?

The token reached a fully diluted valuation above $2.7 billion in early January 2025, at a price around $2.47, after Marc Andreessen’s own social-media shoutout and the broader AI-agent token mania of that period. It has since fallen by well over 99%. As of July 2026 it traded for a fraction of a cent with a market capitalization of roughly $4 million, and it set a fresh all-time low just days before this article was published, according to CoinGecko.

Is ElizaOS legitimate, or was it a scam?

The open-source framework itself is real, actively maintained software used by outside developers independently of any single token. Separately, a federal class-action lawsuit filed in April 2026 alleges the ai16z project misrepresented how autonomous its flagship trading agent actually was and diluted token holders through an undisclosed token migration. Those allegations have not been proven in court, and no defendant has been found liable for anything. Whether the technology works and whether the token and DAO around it were honestly marketed are legally and factually separate questions.

How is the Eliza framework different from Virtuals Protocol or LangChain?

LangChain is a much larger, general-purpose framework for building LLM applications with no native token or crypto focus. Virtuals Protocol is a no-code launchpad, mainly on Base, where users tokenize and co-own AI agents through its own built-in economic layer. Eliza sits between the two: open-source and developer-first like LangChain, but built specifically for agents that hold wallets and act across both social platforms and blockchains, without Virtuals’ built-in tokenization and launchpad mechanics. Which one is the right fit depends entirely on the goal: a general enterprise LLM application, a tokenized agent economy, or a self-hosted bot with native on-chain capabilities each point toward a different one of the three.

Written by the HOGE Wire ai-crypto desk.