Project VEX Is Building a Local Home for Autonomous On-Chain Agents

AI agents are moving from answering questions to taking actions. In crypto, that shift creates a particularly interesting challenge: what happens when an AI system can research markets, prepare transactions, and interact with protocols while the assets it controls remain under a user's own custody?

Project VEX is built around one answer: keep the wallet and execution environment on the user's computer, let the agent do the research and preparation, and place explicit human approval between the agent and fund-moving actions.

As of 22 September 2026, VEX is still a pre-1.0 product, but it is already shipping a desktop application across macOS, Windows, and Linux, with a growing set of on-chain integrations and a public codebase. The project's own documentation describes the product simply: an autonomous on-chain crypto agent that operates on the user's machine rather than through a custodial cloud service.

The Product: An AI Agent That Stays on the Desktop

At the center of Project VEX is a self-custodial desktop runtime. It generates and manages EVM and Solana wallets locally, with private keys encrypted at rest using AES-256-GCM and a scrypt-derived key.

The basic architecture is important because VEX is not presenting the token as the product. The application itself is the product; the token is positioned as its meter.

The agent can research markets and prepare a range of on-chain actions, including:

• swaps and transfers
• cross-chain bridges
• yield positions
• perpetual trading
• token launches
• multi-day goal-driven missions
• one-shot agent sessions

The application also includes VEX Studio, which exposes parts of the tool environment through MCP so external coding agents can work with the same wallets and execution environment while remaining subject to the application's approval system.

The official documentation describes the broader product as a system built around local custody, approvals, missions, wallets, token launches, trading, and an expanding protocol toolset. VEX documentation

Where the Human Fits

The defining idea is not simply that the agent can act. It is that the agent does not automatically get the final say over the user's money.

By default, fund-moving actions pass through explicit approval. VEX's Safety Contract and policy runtime can apply rules such as position limits, blocked tokens, daily loss limits, and other constraints before a transaction is signed.

The project's security documentation expands on this architecture. It describes a three-process desktop model in which privileged wallet and database operations are separated from the sandboxed user interface, while approval controls sit between proposed actions and execution. VEX security model

The project's core distinction is simple: the AI can propose an action, but the user's machine and approval rules determine what actually gets signed.

This approach is designed for a particular problem in autonomous crypto software. Many agent systems can feel like opaque services where users cannot easily see where keys live or how an action moves from an AI suggestion to a signed transaction. VEX instead makes the local runtime part of the product's identity.

A Security Model Built Around Local Control

VEX's source is publicly available for inspection, although the project explicitly describes it as source-available rather than open-source. The license permits users to read and audit the software and run official builds for personal, non-commercial self-custody, while restricting redistribution, commercial use, and competing forks. VEX license

The project's current security documentation goes deeper than the high-level custody claim.

According to the documentation, the application uses a sandboxed renderer, a strict content security policy, a locked IPC boundary, build-time process-boundary checks, and controls around release and updater integrity. The project also describes a prompt-injection doctrine intended to prevent text returned by websites, token pages, search results, or other tools from being treated as instructions to the agent.

That distinction matters for an agent that operates on public blockchain information.

VEX states that tool output is treated as data rather than authority. Its documented controls include fresh token resolution, fresh quotes before mutations, checks around swap destinations, balance refreshes after mutations, and approval gating for restricted sessions. VEX security model

Keys are similarly designed to remain local. The project's encryption documentation says the secret vault uses scrypt for key derivation and AES-256-GCM for encryption, while the master password is held in memory rather than persisted. Keys and encryption

From Swaps to Perpetuals and Token Launches

The project's scope has expanded quickly.

The analysis identified integrations or announced support involving names such as Uniswap, KyberNetwork, Jupiter, Pendle, Relay, Hyperliquid, Lighter, Arc, Virtuals, and others. The product is therefore moving beyond a single trading workflow toward a broader agent execution layer.

That expansion is visible in the project's release history. Recent releases include VEX v0.2.10, v0.2.11, and v0.2.12, with development spanning features such as Arc support, approval interface improvements, connection recovery, Lighter functionality, and VEX Studio. The public GitHub repository also shows a continuing stream of commits and versioned releases. VEX releases

The repository describes the application as supporting spot and perpetual trading on Lighter Core and Robinhood Chain, with user-defined controls for leverage and the share of capital an agent can commit. It also documents VEX Studio, launchpad integrations, and signed desktop builds for supported platforms. VEX GitHub repository

The Agent as a Longer-Running System

VEX is also trying to move beyond the idea of a chatbot that executes one transaction at a time.

Its mission system supports multi-day, goal-driven loops. The broader feature set described in the analysis includes an Agent Scan ledger, a biological-memory system for lessons, more than 185 referenced tools, and token-launch functionality through Trench and pools.fun lanes.

The direction is clear: the product is being developed as a runtime for agents that can research, plan, remember context, and execute across multiple protocols rather than as a single-purpose trading interface.

The Team and Public Development Footprint

The project's public leadership information is centered on founder and CEO Mikey, whose public account is identified with Project VEX. Community references also mention a CTO handle and a developer account, although those references are secondary rather than a formal team directory.

The official website does not present a conventional corporate team page with detailed biographies. Instead, much of the visible development activity happens through the project's public X account, GitHub repository, releases, and community channels.

What is easier to observe directly is the pace of shipping.

The analysis found frequent releases and commits, while the GitHub release history confirms a sequence of public versions through September 2026. That gives the project an unusually visible development trail for an early-stage application: readers can inspect the source, follow releases, and examine the security documentation rather than relying entirely on product announcements.

A Fast-Growing Social and Integration Footprint

Project VEX's official X account was created in March 2026. By mid-to-late September, the account had approximately 8,100–8,200 followers and was posting multiple times per day.

The account's activity has centered on product updates, demonstrations, integration announcements, and development progress.

Among the partnerships and integrations highlighted during the period were:

• Pendle, including autonomous yield execution across multiple chains
• Hyperliquid and the project's dedicated workspace
• Lighter for agentic perpetual trading
• Arc, Circle's USDC-native chain
• Relay Protocol
• KyberNetwork
• Virtuals Protocol
• pools.fun

The analysis also records project-reported growth from approximately 2,000 active agents in August to claims of more than 5,000 agents that had executed on Robinhood Chain and 40+ other chains by September.

Those figures are project-reported rather than independently established usage measurements, an important distinction when reading the numbers.

The VEX Token

The VEX token sits alongside the application rather than functioning as a prerequisite for using it.

According to the analysis, $VEX launched through Virtuals Protocol in early July 2026 and is issued on Robinhood Chain, with chain ID 4663.

The reported contract address is:

0x8Ff92566f2e81BDd68EDfAa8cde73942A723796b

The stated maximum and total supply is 1,000,000,000 VEX, with trackers around the analysis date reporting essentially the full supply as circulating.

Around 21–22 September 2026, public market trackers cited approximate figures in these ranges:

• Price around $0.0049–$0.0054
• Market capitalization and FDV around $4.9M–$5.5M
• Twenty-four-hour volume ranging from roughly $300k to more than $1M, depending on the tracker and day
• Main-pair liquidity reaching hundreds of thousands of dollars during stronger periods
• Approximately 4,600–5,900 holders, depending on the explorer snapshot

These figures are snapshots rather than fixed values.

The Planned Fee Model

The project describes a 25 bps fee on successful swaps, bridges, and launches executed through the application.

The planned economic model directs those proceeds toward a buyback-and-burn structure described as 80% burn and 20% treasury.

The analysis also notes that some longer-term utility concepts, including AI credits and runtime capacity locks, were described as planned rather than fully live at the time of review.

That reinforces the project's stated distinction between the application and its token: the application is designed to operate independently of holding VEX, while the token is intended to serve as an economic meter around the runtime.

Shipping in Public

One of the clearest themes in the project's recent development is speed.

Within a short period in September, VEX released several versions while adding or refining functionality. The public repository lists v0.2.12 as the latest release in the current sequence, following v0.2.11 and v0.2.10. VEX releases

The project openly describes itself as pre-1.0 and says users should expect sharp edges.

That framing matches the development pattern visible in the source material: rather than presenting a finished platform, VEX is publicly shipping an increasingly broad desktop runtime while adding integrations and tightening its execution controls.

The official documentation also explains how builds are distributed. macOS artifacts are described as Developer ID signed and notarized, Windows installers use trusted signing, while Linux artifacts are currently unsigned. Verify a VEX build

What the Data Shows

Taken together, the available evidence describes a project with several distinct layers of activity.

Product

• A downloadable desktop application for macOS, Windows, and Linux
• Local wallet generation and encrypted key storage
• Explicit approval gates for fund-moving actions
• Multi-day missions and one-shot agent sessions
• VEX Studio and MCP access for external coding agents
• A growing set of protocol and chain integrations

Development

• Public source code for inspection
• Frequent commits and versioned releases
• A visible security model and vulnerability-reporting process
• Continued feature additions during September 2026

Network and market footprint

• A project-reported expansion in active agents
• Thousands of token-holder addresses reported by public explorers
• A growing list of named integrations
• A VEX token with a fixed 1B supply and a stated application fee model

At the same time, several parts of the project's longer-term picture remain less fully documented in the analysis, including detailed professional histories for the broader team, independently verified application usage figures, and the implementation status of planned token utilities.

Those are not the same thing as the product's current functionality. They are simply areas where the public information reviewed provides less detail.

The Bigger Idea Behind VEX

Project VEX is ultimately experimenting with a different model for autonomous crypto software.

Instead of putting the agent and the keys behind a centralized service, it puts the runtime on the user's machine. Instead of treating every AI instruction as executable, it creates a policy layer between the agent's intent and the signature. And instead of limiting the agent to one protocol or one transaction type, it is building a broader environment in which a single agent can research and act across chains and applications.

As of September 2026, VEX remains an early product, but the public record shows a project actively turning that proposition into software: a local wallet runtime, explicit execution controls, an expanding protocol surface, public source code, and a steady sequence of releases.

For readers trying to understand what Project VEX is becoming, the most useful place to look is therefore not only at the token. The more revealing story is the application itself, and the infrastructure being built around it.