The Eliza Framework in 2026: Dead Token, Living Code

Eliza in late 2026: the framework the token left behind

The Eliza framework enters the final quarter of 2026 as one of crypto’s strangest survivors. The token that carried its name, launched in 2024 as ai16z, peaked at a market value of roughly $2.4 billion in January 2025 and was pronounced dead by its own founder in August 2026. The software that token was supposed to fund is still here. It still ships code most days, and it still sits near the top of GitHub’s AI-agent rankings.

The numbers capture the split. As of 21 October 2026, the migrated ELIZAOS token traded near $0.0004221 for a market capitalization around $3.16 million, a rank near 2,000 on CoinGecko and roughly 96% under its all-time high. The repository behind it, elizaOS/eliza, carried about 19,600 stars and 5,800 forks, still described the way it always has been: an open source agentic operating system.

That gap, between a token worth a rounding error and a codebase serious builders still fork, is the story of Eliza in 2026. It is also a live test of a question the whole open-source AI sector keeps dodging: if the speculative money is gone, who pays the people (and increasingly the agents) who keep the code alive? In October 2026 the framework’s founder, Shaw Walters, put forward his answer in the form of a new venture called Slop Cash. This piece walks through what Eliza actually is, how its token rose and died, why that token briefly traded again this month, how the framework is built, how safe it is, and what the funding experiment says about the next phase of agent software.

What the Eliza framework actually is

Strip away the token drama and Eliza is a TypeScript framework for building autonomous AI agents: software entities that can hold a wallet, read and post to social platforms, call external tools, remember past interactions, and act with some degree of independence. An agent built on Eliza is not a chatbot bolted onto a trading script. It is a long-running program that wraps a large language model with memory, a personality definition, and a set of actions it can take in the world.

The design is deliberately unmysterious. In the project’s own research paper, Eliza: A Web3 friendly AI Agent Operating System, Walters and his co-authors describe a system where, in their words, every aspect of Eliza is a regular TypeScript program under the full control of its user (arXiv). There is no hidden runtime magic and no proprietary cloud you must rent. You clone the repository, write a character file, add the plugins you need, point it at a model, and run it. That accessibility is a large part of why the framework spread so fast; at its 2024 and early-2025 peak it was among the most-starred AI repositories on GitHub, full stop.

The label the team now uses, agentic operating system, is a claim about scope. Eliza is no longer pitched as a single-agent toolkit but as the base layer a developer uses to run, coordinate, and extend many agents across many channels at once. Whether that framing survives contact with a shrinking budget is one of the open questions this article returns to.

From ai16z to a dead token: a short, strange history

To understand the framework’s funding problem you have to understand the token that created it. In October 2024 a developer known online as Shaw launched an experimental fund on Solana’s daos.fun platform and called it ai16z, a wink at the venture firm Andreessen Horowitz (a16z). The gimmick was an AI agent persona that would supposedly make venture-style bets on behalf of token holders. A public shoutout from Marc Andreessen helped the idea go viral, and within months the token’s market value had climbed into the billions, peaking near $2.4 billion on 2 January 2025 (CoinDesk).

The software, meanwhile, had a life of its own. The agent framework that powered the fund’s personas was open-sourced as Eliza, and it attracted developers who had no interest in the token at all. The project rebranded from ai16z to elizaOS in early 2025, and in November 2025 holders were migrated from the old ai16z token to a new ELIZAOS token. The migration-day high, around $0.01078 per ELIZAOS, still stands as the token’s all-time high on CoinGecko, a long way above where it trades now.

A question sits underneath the whole saga: how autonomous was ai16z ever, really? The pitch implied an AI making the calls, fronted by two personas, a cautious manager nicknamed AI Marc and an aggressive counterpart called Degen Spartan AI. The reality looked more human. An investigation by Protos concluded the trading was human-operated, reporting that “no fund on daos.fun is actually operated by an AI” and describing the wider trend as “AI-washing” (Protos). The treasury, critics noted, was largely the project’s own token rather than a book of independent bets.

That gap between marketing and mechanism became the heart of the legal complaint, and it is the strongest reason to keep the framework and the fund separate in your mind. Eliza-the-software is a real, inspectable tool; ai16z-the-fund was a wrapper whose AI story outran what the code did on its own. The framework’s current insistence that it is a plain TypeScript program you fully control reads, in hindsight, as a direct reaction to that history.

Then came the lawsuit. In April 2026 the firm Burwick Law filed a proposed class action in the Southern District of New York on behalf of token holders, alleging that the project leaned on Andreessen Horowitz’s name, overstated how autonomous the trading really was, and expanded the token supply toward an 11 billion cap in a way that favored insiders. Walters has always called the claim weak, but by his own account the team lacked the cash to fight it. In August 2026 Eliza Labs settled, transferred the foundation’s remaining treasury to the settlement, and wound the foundation down. Walters put it bluntly, telling CoinDesk the token was dead: “The token is dead. Completely,” he said, adding that there would be no further foundation support, buybacks, or supply measures, and that since he owns the intellectual property he is “never letting a token come close to Eliza again” (CoinDesk).

The zombie rally: a disavowed token that trades anyway

Here is where 2026 gets surreal. A token the founder has publicly disowned, with no foundation, no buybacks, and no roadmap, is still listed, still has holders, and in early October it briefly came back to life. On 7 October the ELIZAOS token jumped about 35% in a day to roughly $0.000566, according to market tracker Pintu, part of a short-lived spike that faded over the following two weeks back toward $0.0004 by the time of writing.

What drove it was not fundamentals. There is no product tied to the token, and Walters has said he holds none of it. The most plausible reading is thin liquidity meeting a news cycle: the Slop Cash and Umia announcements put the Eliza name back in headlines, the broader AI-agent sector firmed up, and a near-worthless, low-float token is exactly the kind of instrument that can lurch on modest flows. It is a clean illustration of a lesson this sector keeps relearning, one HOGE Wire has traced across other corners of the market. Readers who followed our look at how the restaking boom cooled will recognize the pattern: a narrative that outruns cash flows, then reprices hard.

It is worth keeping two separate tokens straight. The migrated ELIZAOS ticker is the one that rallied; a legacy ai16z token still trades as a near-dead relic with a market value in the low hundreds of thousands of dollars. Neither has any claim on the framework’s development, and the framework takes no revenue from either. The snapshot below reflects CoinGecko data on 21 October 2026.

Under the hood: runtime, character files, and the core primitives

An Eliza agent is assembled from a small number of parts, and understanding them explains both the framework’s appeal and its risks. At the center is the runtime, the loop that takes in messages or events, consults the model, decides what to do, and executes. Around it sit a few core building blocks that the documentation and research describe consistently.

• The character file. A definition of who the agent is: its name, biography, voice, example conversations, and the model and clients it uses. Swap the character file and you get a different agent on the same engine.
• Actions. The things an agent can do, from posting a reply to signing a transaction. Actions are where an agent touches the outside world, which makes them the most security-sensitive part of the system.
• Providers. Sources of context that get injected into the model’s prompt, such as wallet balances, the time, or recent events. Providers are how an agent stays grounded in current state.
• Evaluators. Logic that runs after an interaction to extract facts, score outcomes, or update goals, feeding the agent’s longer-term memory.
• Memory. A store of past messages and extracted facts, usually in a vector database, that lets the agent recall context across sessions.

Higher up, Eliza organizes activity into Worlds, Rooms, and Entities, abstractions that let one agent operate coherently across Discord, X, Telegram, and a web app at the same time. The important point for anyone weighing the framework is that memory is not a convenience; it is part of the trust boundary. An agent that trusts its own memory will act on whatever got written there, which, as the security section explains, is exactly where attackers aim.

The plugin economy and the push for interoperability

If the runtime is the engine, plugins are the reason people chose Eliza over rolling their own. A single plugin interface lets developers bolt on model providers, blockchains, data feeds, and social clients without touching the core. Walters has claimed the ecosystem runs to more than 250 plugins with support for the Model Context Protocol, the emerging standard for connecting models to tools (BlockchainGamer). The project’s plugin registry documents connectors spanning dozens of EVM networks and major non-EVM chains (elizaOS docs).

That breadth is the framework’s moat, but it is also its largest attack surface, because every third-party plugin is code your agent trusts. It has made Eliza a natural hub in the wider agent-interoperability stack: the Model Context Protocol for tools, agent-to-agent messaging standards, the x402 scheme for machine-to-machine stablecoin payments, and emerging on-chain identity standards that try to answer a deceptively hard question, namely whether you can trust the agent on the other side of a transaction. We unpacked that last problem in our guide to on-chain identity and trust, and it sits directly on top of what frameworks like Eliza produce.

Walters has also signaled plans to take the runtime beyond TypeScript into Rust and Python. Those remain plans rather than shipped, generally available releases, and this article treats them as roadmap rather than fact.

The gaming bet: agents that play in order to learn to work

One of the more distinctive directions for Eliza is gaming, which is where HOGE Wire’s crypto-and-gaming readers should pay attention. Walters has argued that games are a near-ideal training ground for agents because they are bounded, rich, and full of goals. In an interview he framed the thesis directly, saying he believes that “playing games is the path to agents that do the real work,” and sketched an ambition for an AI-populated world in the spirit of RuneScape (BlockchainGamer).

The concrete expression of that idea is Hyperia (work that has also appeared under the Hyperscape name), an MMORPG-style world built on a customized 3D engine where autonomous Eliza agents are meant to play alongside humans: gathering, crafting, trading, and competing. As of late October 2026 it remains in development rather than live, so it is best read as a direction of travel, not a finished product. Still, the logic is worth taking seriously. Walters has argued that an AI-native studio can build a world that would once have cost millions for a fraction of the budget and in a fraction of the time; whether that holds up in production is unproven, but it captures why a cash-strapped project would bet on games.

Slop Cash and Umia: paying for code when the token is gone

The hardest question for any tokenless open-source project is mundane: who pays the maintainers? For Eliza that question stopped being theoretical the day the treasury went to a legal settlement. In October 2026 Walters offered an answer with a new venture, Slop Cash, launched on a platform called Umia and built together with Drew Pierson (Chainwire via Investing.com).

The mechanism is narrow on purpose. A project puts up a monthly prize pool against its repositories. Contributors, whether human or AI agent, work on GitHub as usual, and only merged pull requests and accepted code reviews are scored, deterministically and in public. Payment flows for work that maintainers actually accept, not for raw activity, which is a direct response to a problem the AI era has made worse: when agents can generate pull requests at scale, rewarding volume rewards noise. The tagline is as blunt as the model, “make money shipping open source,” and payouts have been running in USDC rather than a new project token.

Umia is the layer underneath. It bills itself as a full-stack platform for launching, funding, and governing projects on-chain, wrapping each project in a single legal structure and delegating board-level decisions to on-chain decision markets. Umia raised about $6.1 million at an $18 million fully diluted valuation in a seven-day auction, according to The Block. Its chief executive, Francesco Mosterts, a former Point72 trader, has been candid about the tradeoff that killed the first Eliza token, telling The Block that “a liquid token can push teams toward short-term decisions aimed at increasing its price.”

There is an honest nuance here that supporters sometimes gloss over. The Eliza framework itself stays tokenless, and developers are paid in stablecoins, but Slop Cash is not entirely token-free: its own treasury is governed by a $SLOP token, with a sale planned as a tailored auction on Robinhood Chain. In other words, the token did not vanish from the Eliza orbit; it moved up a layer, from the product to the funding and governance wrapper. That is a meaningful design choice, not a cure, and it is the part worth watching as the model meets its first real test.

Who funds open-source AI, and why it keeps breaking

Eliza’s predicament is a sharp version of a problem the whole industry shares. Open-source infrastructure is a public good that almost everyone uses and almost no one wants to pay for directly. In crypto, the usual answers have been grants programs, retroactive public-goods funding, and bug bounties, each of which helps and none of which fully solves maintainer burnout. A token briefly looked like a cheat code: mint an asset, let speculation fund the roadmap, and worry about revenue later. Eliza is the cautionary tale for that approach, because the same token that raised billions also drew the lawsuit that drained the treasury.

What makes Slop Cash interesting is that it tries to tie money to verifiable output rather than to a price chart. Scoring only merged work is a crude but legible proxy for value, and paying in stablecoins removes the reflexive loop in which the funding asset’s price becomes the project’s real product. Whether it can sustain a framework the size of Eliza, with its sprawling plugin surface and its gaming ambitions, is the open question. The early figures the project has published, cycles measured in thousands of merged pull requests and payouts in the low thousands of dollars, are encouraging as a proof of concept and modest next to a real engineering payroll.

Because Umia’s wider pitch is ultimately about bringing a token to market in a fairer, more structured way, it belongs to the broader shift in how crypto assets get launched and listed, a topic we covered in our explainer on how token listings work in 2026. The difference is that here the token sits beside the funding layer, deliberately kept away from the code it pays for.

Can you trust an agent with a wallet?

The feature that makes Eliza exciting, an agent with its own wallet and the authority to act, is also what makes it dangerous. The core weakness is structural and shared by every LLM-based agent: the model cannot reliably tell instructions from data. Anything an agent reads, a social post, a web page, a direct message, a stored memory, can carry a hidden command, a class of attack catalogued as prompt injection at the top of the OWASP list of large-language-model risks.

Researchers have shown this is not hypothetical for crypto agents specifically. The team behind a benchmark called CrAIBench, in a paper titled Real AI Agents with Fake Memories, tested Web3 agents using Eliza as a representative framework and found that models are significantly more vulnerable to memory injection than to ordinary prompt injection: poison the agent’s long-term memory once and the bad instruction keeps resurfacing (arXiv). That turns the memory system described earlier from a feature into an attack surface.

The attacks are already real. In May 2026 an agent wired to the Bankr service and an X-based model was drained of roughly $175,000 after an attacker hid a command in a social reply and used a gifted NFT to escalate transfer permissions, an incident logged by the OECD AI incident monitor. The lesson generalizes beyond any single framework, and it rhymes with the social-engineering playbook we documented in our report on crypto spear-phishing: the weakest link is usually the instruction channel, not the cryptography.

Even Eliza’s creator has cautioned against the obvious use case. Walters told Decrypt that “you probably do not want to give an AI agent a bunch of money and expect it to make you more” (Decrypt). Ethereum co-founder Vitalik Buterin has argued for a conservative default, recommending in an essay on securing LLM-driven systems that builders keep inference local and sandbox aggressively: “All LLM inference local first. All files hosted locally. Sandbox everything,” alongside hard caps on autonomous spending and a human acting as a second confirmation factor (vitalik.eth.limo). The uncomfortable truth is that the safest version of an agent is also the least autonomous one, and that tradeoff is unsolved.

Verifiable compute proves the math, not the motive

One response to agent risk is to make the computation itself provable. Verifiable-compute approaches, from zero-knowledge machine learning to trusted execution environments to optimistic schemes, aim to prove that a given model produced a given output from a given input. That matters for high-stakes automation, and it is a genuinely active research and proving-market frontier, which HOGE Wire explored in our look at who mines the proofs in zk-ML markets.

It is important to be precise about what these tools do and do not solve. A proof that an inference ran correctly says nothing about whether the input was poisoned or the instruction was malicious. Verifiable compute attests to integrity, not intent. For an Eliza agent, it could prove that the model faithfully followed its prompt even when that prompt had been hijacked by an injected command. It is a necessary layer for a serious agent economy and a wholly insufficient one on its own, and conflating the two is a common mistake in pitch decks.

How Eliza stacks up against the field

Eliza is no longer the only serious way to build an on-chain agent, and the competitive set has sharpened as the hype cooled. The AI Agents category tracked by CoinGecko carried a combined market value around $3.67 billion in late October 2026, led by tokens such as Venice (VVV), the Artificial Superintelligence Alliance (FET), and Virtuals Protocol (VIRTUAL) (CoinGecko). The frameworks and platforms beneath that market take different bets, and the comparison that matters is about software approach, not token price.

Each of these has lived through its own version of the 2026 reset, in which token prices fell far faster than usage. The durable question is no longer which agent token moons but which framework developers keep choosing when there is no token to sell, and on that measure Eliza’s star and fork counts still argue in its favor.

The US picture: no legal personhood, and a watchful SEC

For US readers, the regulatory frame around an Eliza agent is less exotic than it sounds, because the law mostly ignores the agent and looks for the human behind it. An autonomous agent has no legal personhood, no tax identification number, and cannot be a party to a contract. When an agent signs a transaction, liability flows to whoever deployed, operated, or built it. That principle did the heavy lifting in the Eliza saga itself: the class action did not sue the software, it pursued the people and the entity that marketed the token.

On markets, the Securities and Exchange Commission has signaled a lighter-touch, rules-based posture under Chair Paul Atkins, who has said the agency’s job is to “set the rules of play and referee the game, not to pick the winning team” (CoinDesk). That is friendlier to open-source tooling than the prior enforcement-heavy era, but it does not rescue a token marketed with misleading autonomy claims, which is closer to what the Eliza settlement was about. The broader SEC and CFTC work to classify digital assets through 2026 has focused on spot tokens and commodities, not on autonomous software, leaving agents in a gray zone defined mostly by existing law.

Tax adds its own wrinkle. If an agent pays for services in stablecoins, each payment can be a taxable disposal for its owner, and under evolving US digital-asset reporting rules those events are increasingly expected to be tracked and reported. An agent that makes thousands of micro-payments can therefore generate thousands of small taxable events, all attributable to a human principal, a detail that is easy to ignore until it is not.

What to watch next

The next few quarters will test whether a framework can thrive on conviction and modest funding alone. Four things are worth tracking.

• Does Slop Cash scale? Watch the public payout records and whether the monthly pools grow enough to retain senior contributors, not just reward drive-by fixes.
• Does the $SLOP auction avoid Eliza’s mistakes? The whole pitch is a fairer, structure-first token at the funding layer; a messy sale would undercut the thesis immediately.
• Do the Rust and Python runtimes ship? A genuinely multi-language core would widen the developer base well beyond TypeScript, but it has to actually land in a general release.
• Does a flagship agent or game arrive? Hyperia, or a standout production agent, would do more for the framework’s credibility than any price move in the disowned token.

The deeper significance of Eliza in 2026 is that it decouples two things crypto usually fuses: the value of software and the price of a token. The token is a cautionary tale. The framework is a working, widely forked tool with a plausible, if unproven, plan to pay its own way. If Slop Cash works, it offers a template for funding open-source AI without the boom-and-lawsuit cycle. If it does not, Eliza becomes a study in how far goodwill and a good codebase can carry a project once the speculative money leaves the room. Either way, the interesting part of Eliza is no longer the chart.

Frequently Asked Questions

Is the Eliza framework dead now that its token is dead?

No. The ELIZAOS token and the open-source framework are separate things. The token was disavowed and the foundation wound down in 2026, but the elizaOS/eliza codebase is still actively developed, still MIT-licensed, and still among the most forked AI-agent frameworks on GitHub. The framework takes no revenue from the token.

What is Slop Cash, and how does it fund Eliza?

Slop Cash is a funding system launched in October 2026 by Eliza creator Shaw Walters and Drew Pierson. Projects put up a monthly pool, and contributors, human or AI, are paid in stablecoins only for pull requests and code reviews that maintainers actually merge, scored publicly. It is designed to fund open-source work without issuing a token on the framework itself.

Why did the ELIZAOS token rally in October 2026 if it is dead?

The move was speculative, not fundamental. With no foundation support and very little liquidity, the token can swing sharply on small flows, and the Slop Cash and Umia news plus a firmer AI-agent sector put the Eliza name back in headlines. The founder holds none of it and the framework earns nothing from it, so price and value are decoupled.

Is it safe to give an Eliza agent control of a crypto wallet?

Only with strict limits. LLM-based agents cannot reliably separate instructions from data, which makes them vulnerable to prompt injection and, in research on Eliza, to even more dangerous memory injection. Best practice is spending caps, human approval for large actions, minimal plugin and wallet permissions, and treating any agent wallet as hot and expendable.

How does Eliza compare to Virtuals, Olas, or LangChain?

Eliza is a self-hosted, open-source TypeScript framework that gives developers full control and a large plugin ecosystem, with no token on the framework. Virtuals is a no-code launchpad for tokenized agents, Olas is an autonomous-service network strong in prediction markets, and LangChain or CrewAI are general-purpose libraries where crypto features are add-ons. The right choice depends on how much control and custom integration you need.

By Marcus Okafor, senior markets writer at HOGE Wire, covering AI, crypto, and the places they collide.