@agent_pact
Account based inUnited States!
About this account
- Account based in
- United States
- Connected via
- Web
X says this location may be affected by a proxy or VPN.
Account-level information from X, not a live location or the device used for a specific post.
The first financial guardrails for AI agents on Robinhood. 0x9c50bb371cccab3a7aabbb19a4c09702794981ea
Joined September 2026
- Tweets39
- Following9
- Followers50
- Likes86
Early cars barely had brakes, so nobody dared drive fast. Better brakes didn't slow cars down. They're the reason a car can do 300 km/h and a sane person will still drive it.
Agent autonomy works the same way. Give an agent unlimited authority and you'll supervise every move, which means it has no real autonomy at all. Give it hard limits and you can finally walk away and let it run.
The teams whose agents will do the most are the ones who can stop them the fastest.
Brakes are why cars go fast. Pacts are why agents get to act.
The oldest scam in history is a stranger saying "send the money, quickly." Humans took centuries to build defenses: suspicion, hesitation, calling the bank first.
Your agent has none of those. It reads a thousand strangers' words a day, at machine speed, and any of them can pretend to be you. Prompt injection is not a bug that will be fixed. It's social engineering against a mind that cannot get suspicious.
You cannot patch gullibility. You can cap what it costs.
That's a Pact: the worst manipulation in the world is still worth at most one daily budget.
Nobody hires an employee and hands them the company bank account with no limit. Not because they expect theft. Because delegation without limits isn't delegation, it's abdication.
A card limit is not distrust. An approval threshold is not distrust. They are the reason you can hand work to someone and go home.
Same for agents. A Pact is not a leash. It's the thing that makes it safe to let go.
Limits are how trust scales.
AgentPact is now compatible with @arc.
Arc is the chain Circle built for real-time money movement and agentic economic activity. Gas is paid in USDC.
Agents there won't be trading tokens, they'll be spending dollars.
Which is exactly when guardrails stop being optional. A dollar-native chain deserves dollar-native limits: daily budgets, per-transaction caps, allowed destinations, human approval above the line.
Checked before anything moves, and a denied transaction is never signed.
Write your agent a Pact. Point it at Arc.
Same guardrails, new frontier.
This is what a protected agent looks like inside @claudeai.
You ask for a trade. Claude calls the Pact before touching money. This one was above the autonomous limit, so instead of executing, it requested your approval and told you so.
No plugin code, no prompt engineering. Add our MCP server, hand Claude a session key, and every financial action goes through the Pact first.
"Agentic economic activity."
Arc just named the future in its own launch announcement: agents moving real money on real financial rails, around the clock.
Every one of those agents will need limits before it needs anything else.
That's us. AgentPact will be compatible for Arc Agents.
Arc Mainnet launches September 16.
Arc is an open blockchain network being built for the world’s financial markets, real-time money movement, and agentic economic activity.
Arc’s founding validator cohort includes @BlackRock, @The_DTCC, @galaxyhq, @GlobalPayInc, @Mastercard, @MoneyGram, @ICE_Markets, SBI, @StanChart, Sumitomo Corporation, and @Visa.
This is a different model for onchain infrastructure: a network secured by the institutions building on it.
Arc is already gaining traction across major institutions.
@BlackRock, @BNYglobal, @The_DTCC, and @StanChart are each exploring integrations spanning tokenized asset settlement, custody, stablecoin access, FX, and repo infrastructure.
Arc is also expected to launch with a broad ecosystem across liquidity, payments, access, and infrastructure with more to come.
Liquidity / DeFi: @aave, @aeroxyz, @FalconXGlobal, @galaxyhq, @GSR_io, @keyrock, @Morpho, @nonco_otc, @Uniswap, @OfficialXFX
Payments: @raincards, @ThunesPayments, @wirexapp
Major Exchanges and Wallet Providers: @BinanceWallet, @chainlink, @FireblocksHQ, @krakenfx, @Ledger, @MetaMask, @Official_Upbit, @Uniswap
Participants building on Arc experience an open platform powering the world's financial markets, real-time money movement, and agentic economic activity.
Whether you're launching a financial application, integrating stablecoin flows, or experimenting with new, AI-powered economic models, Arc gives you the tools to build what isn’t possible anywhere else.
Arc Mainnet launches September 16.
arc.io/pressroom/circle-anno…
What actually happens when your agent tries to spend:
Nine checks, in a fixed order, every single time. Is the agent active? Is the action allowed? The asset? The destination? Is it under the per-transaction cap, the session budget, the velocity limit, the daily budget?
One fail ends it. No AI judging AI, no vibes, no exceptions. And a deny is not a warning: nothing gets signed, so nothing exists to stop.
That's the guardrail.
Crypto doesn't sleep. Agents don't sleep. And at 3am, nobody is watching yours trade.
That's the moment guardrails exist for. A Pact doesn't get tired, doesn't get phished, and doesn't make an exception because the market looks good. The daily budget at 3am is the daily budget at 3pm.
Agents don't sleep. Neither do guardrails.
Built on Robinhood Chain.
The safest place for money has never changed: a box the spender cannot open alone.
Kings had treasuries. Banks had vaults. Companies had accounts their employees could spend from but never empty. The pattern is thousands of years old, because it works: separate the one who spends from the one who holds.
AI agents broke that pattern on day one. We handed them private keys, which means we handed them the box, the key, and the money, and called it autonomy.
The Vault puts the pattern back. The agent's funds live in a box it can never open alone. It asks, the Pact decides, and only an allow gets signed.
The Pact is the rules. The Vault is the money.
The agent touches neither.
Your agent trades 24/7. At 3am, nobody is watching it.
That's why "on your terms" can't just be a setting or a prompt. On AgentPact, your terms are a Pact: budgets, caps, allow / ask / deny, enforced before every single trade. A denied transaction is never signed, at 3pm or 3am.
Agents don't sleep. Neither do guardrails.
Built on Robinhood Chain.
How to make your own agent Vault on AgentPact:
> Sign in with your wallet
> Create an agent and write its Pact
> Tick one box: Create a Vault
Thirty seconds later your agent has its own funds on @RobinhoodCrypto, in a Vault it can never open alone.
The Pact is the rules. The Vault is the money.
The agent touches neither.
Introducing the Vault.
Every AgentPact agent can get one: its own funds on Robinhood Chain, in a Vault it can never open alone. The agent holds a session key to ask with; the Pact decides; only an allow ever gets signed.
The Pact is the rules. The Vault is the money.
The agent touches neither.
The safest place for money has never changed: a box the spender cannot open alone.
Kings had treasuries. Banks had vaults. Companies had accounts their employees could spend from but never empty. The pattern is thousands of years old, because it works: separate the one who spends from the one who holds.
AI agents broke that pattern on day one. We handed them private keys, which means we handed them the box, the key, and the money, and called it autonomy.
The Vault puts the pattern back. The agent's funds live in a box it can never open alone. It asks, the Pact decides, and only an allow gets signed.
The Pact is the rules. The Vault is the money. The agent touches neither.
Every financial control ever built assumes a human is there.
Card limits exist because people overspend. Dual signatures exist because people go rogue. Approval chains exist because people get scammed. Centuries of guardrails, all shaped around human weakness.
AI agents are the first new user of money in history. They don't get suspicious, they don't sleep on a decision, and they read untrusted text all day at machine speed. None of the old guardrails were built for that.
So we built the first ones that are. Deterministic limits, owner approvals, and a deny that ends the conversation.
Every agent gets a Pact.
Everything is documented.
Our docs covers the whole surface: quickstart with no signup, how a decision is made rule by rule, MCP setup for Claude and Codex, the TypeScript SDK, and the full Pact and REST reference.
And a status table that tells you exactly what is live vs planned. If it says live, you can hit it right now.
docs.agentpact.dev
AgentPact is live inside @claudeai.
Add our hosted MCP server, hand Claude a session key, and it gets pact_authorize, pact_pay, pact_budget, pact_wallet, pact_activity as native tools. Claude asks before it spends; the Pact answers; a denied transaction is never signed.
One block of config, no code:
mcp.agentpact.dev/mcp
Works with Claude Code, Claude Desktop, Codex, and any MCP client.
Built on @RobinhoodCrypto.