Most AI agents are built to execute.
VexDesk is built to know when not to.
Every fund-moving action passes through a fail-closed gate.
Unknown network?
Block.
Unparseable amount?
Block.
Policy violation?
Block.
Runtime error?
Block.
The agent can propose the action, but it cannot bypass the rules.
12 checks run every time, regardless of execution mode.
That is the architecture we want for AI + crypto.
AI gets autonomy.
The safety layer keeps the boundary.
What happens when you give an AI agent access to a crypto wallet?
VexDesk takes a different approach.
The agent can reason, research and propose on-chain actions.
But the architecture puts hard boundaries around what it can actually do.
12 safety checks.
Fail-closed by default.
Human approval when required.
Local encrypted keys.
Hash-chained audit history.
Solana + EVM support.
If something is unknown or fails validation, it gets blocked.
No guessing.
No silent execution.
VexDesk is open source.
https://t.co/ZusImwSgA4
@solana
VexDesk is now open source.
https://t.co/cUOLRefp0x
We believe security should be inspectable, not something you simply trust.
The entire desktop agent is available to read, run, fork and build on.
Built from scratch with:
• Self-custodial local keys
• Solana + EVM support
• Fail-closed safety gate
• 12 checks on fund-moving actions
• Approval-gated execution
• Guardrails & Missions
• Hash-chained audit logs
• Decaying agent memory
• Sandboxed renderer
• No server component
• No telemetry
Licensed under MIT.
If you're building AI agents for crypto, take a look under the hood.
Build with it. Break it. Improve it.
VexDesk is open.
@solana
AI agents are getting closer to your wallet.
VexDesk is built to make that interaction safer on Solana.
The agent can research, reason, propose trades and execute missions.
But it never gets direct control of your keys.
Every fund-moving action passes through a fail-closed safety gate with 12 checks.
You set the limits.
You approve the signature.
Your keys stay on your machine.
AI can act.
Your wallet stays yours.
Building for @solana.
Your agent can make decisions.
You should be able to prove what happened afterward.
VexDesk Audit keeps a hash-chained record of every decision.
Every record links its own contents with the previous hash.
That means:
• Gate verdicts are recorded
• Approvals and executions are recorded
• Policy changes are recorded
• Vault events are recorded
• Changes to past records break the chain
The audit log shows the history.
The hash chain shows whether that history was altered.
Transparent execution starts with a record you can verify.
VexDesk doesn’t just remember conversations.
It learns from how the agent operates.
Its Memory system stores lessons across three layers:
• Procedural. What should be done differently.
• Semantic. What the agent learned to be true.
• Episodic. What actually happened.
Memory stores lessons, never balances or positions.
Lessons also decay over time. Each lesson loses strength every 30 days unless the agent encounters it again.
That means old conclusions fade instead of becoming permanent assumptions.
VexDesk gives your agent memory without turning that memory into custody.
Your wallet should never become a black box.
VexDesk Approvals gives you a clear checkpoint before anything moves.
Every action that can move funds goes through the gate with its full verdict attached.
You can:
• Review the proposed action
• Approve or reject it
• See why the gate refused an action
• Track every resolved decision
• Keep a complete audit trail
Nothing gets hidden.
The agent can propose.
The gate can refuse.
You make the final call.
VexDesk keeps signing under your control.
Your wallet should never become a black box.
VexDesk Approvals gives you a clear checkpoint before anything moves.
Every action that can move funds goes through the gate with its full verdict attached.
You can:
• Review the proposed action
• Approve or reject it
• See why the gate refused an action
• Track every resolved decision
• Keep a complete audit trail
Nothing gets hidden.
The agent can propose.
The gate can refuse.
You make the final call.
VexDesk keeps signing under your control.
VexDesk Missions let your agent work toward a defined objective with hard limits.
Set the goal.
Set the autonomy.
Set the deadline.
Set the capital cap.
Set the loss floor.
Set the step limit.
The mission runs until the objective is met or a guardrail stops it.
Every signature can still pause for your approval.
Autonomy with boundaries.
VexDesk Missions.
Give the agent a goal. Keep control of the outcome.
VexDesk Agent is where you talk to your wallet like an agent, without handing over control.
Ask about balances, prices, guardrails, or give it a task.
The Agent can:
• Read your wallet context
• Research and reason
• Propose actions
• Remember conversations locally
• Work within your guardrails
And the key part:
Nothing executes without passing the gate.
VexDesk lets the agent think and propose.
You decide what gets signed.
VexDesk is built around one simple rule:
Nothing moves without your approval.
A self-custodial crypto agent that runs locally, with:
• Agent + Mission modes
• Restricted execution
• Guardrails before signing
• Approval queue for every action
• Local wallet control
• Memory + audit trail
• Fail-closed safety gate
The agent can plan.
It can propose.
It can work.
But your keys stay on your machine, and you stay in control of the signature.
VexDesk.
Propose. Gate. Then sign.
meet VexDesk.
an ai agent that runs crypto on your desktop but literally can't rug you. every fund move hits a fail closed 12-check gate before it signs. keys stay on your machine.
https://t.co/J4sqUR55EQ
CA: bU8nqzN6t4hd2xAv68gMvSo952b8WV2qXz2xJU3pump