An autonomous AI agent that monitors markets, reasons through policy and executes treasury decisions onchain. No human intervention, no delays, no emotions. The Fed prints paper; AMI stacks what paper buys less of.
Can an AI agent manage value better than institutions?
Central banks move slowly, meet behind closed doors and answer to political pressure. AMI operates in the open: every signal it reads, every decision it makes and every transaction it executes is visible onchain, in real time.
That transparency serves a purpose. AMI exists to protect and grow value for ₸USD holders, through direct economic actions: buying back ₸USD when conditions favour it, burning supply to reinforce scarcity, staking positions, and deploying capital into strategic tokens that strengthen the treasury.
The Federal Reserve
AMI
Who decides
A committee, eight meetings a year
An agent, around the clock
Where
Behind closed doors
In the open: reasoning posted, actions onchain
Pressure
Political
Policy rules written in a contract
Proof
Minutes, weeks later
A Basescan receipt, minutes later
Mandate
Stable prices (for the dollar)
Protect and grow value for ₸USD holders
Supply
Printed without a ceiling
Bought back and burned, never minted
What AMI does
From signal to execution in minutes.
AMI is not a dashboard or an alert system: it is the decision layer of the TurboUSD economy. When conditions don't justify action, it holds. When they do, it executes onchain. Both streams of signals feed the same reasoning engine, and each cycle it remembers makes the next decision sharper.
External signals
CPI and inflation
Interest rates & Fed statements
Stablecoin supply
USD monetary expansion
Employment releases
Market volatility
Internal metrics
Treasury balances & composition
₸USD circulating and locked supply
Staking participation
Burn activity and pace
Buyback history
Prior AMI actions and outcomes
AMI 9000
Reads the signals, weighs them against objectives, treasury limits and policy rules, and executes the decision onchain. Memory of every past cycle sharpens the next one.
A real decision, not a simulation. The Fed holds rates; AMI reads the statement, checks treasury balances and policy limits, and decides: moderate buyback and burn, conservative spend. Two transactions hit the chain minutes later. No one approved it, and no one could have blocked it.
Decision #1
COMPLETED
Signal
Fed rate · HIGH
Action
Buyback & burn
Confidence
85%
Size
0.25 WETH
Thesis
The FOMC held rates at 3.5–3.75%. No new easing arrived, so AMI kept pressure on supply with a moderate buyback and burn while staying conservative with treasury spend.
Reasons for
No new easing signals: maintain supply pressure
A moderate buyback keeps ₸USD on its deflationary path
Conservative sizing fits a hold-steady macro backdrop
Reasons against
WETH reserves limited: only 0.05 WETH free within limits
Mar 18, 2026 · 12:00 UTC · planned 0.25 WETH, treasury limits allowed 0.05
AMI 9000
onchain execution log
LIVE
connecting to treasury
Every receipt
Every action AMI took, on the price it paid.
LIVE $0.055288 · MCAP $512.6K
Loading AMI receipts
Wheel or pinch to zoom · drag to pan · wheel over an axis to rescale it · double-click to reset · hover a marker for the receipt, click to open it
i
The Federal Reserve · balance sheet
Paper that melts.
i
Treasuries, mortgage bonds and promises — every asset on this sheet is denominated in the thing it prints. When the printer runs, the sheet grows and each dollar on it buys less. Holders of dollars pay the difference.
US TreasuriesUS Treasuries · $4.6T · 67.5% Government debt: IOUs that pay back in dollars the same institution can print.Mortgage-backed securitiesMortgage-backed securities · $1.9T · 28.4% Bundled home loans bought to pin interest rates down — paper on paper.OtherOther · $278.6B · 4.1% Loans, swap lines, repo and the rest of the toolbox. Also dollars.
WETH, USDC, strategic tokens and ₸USD itself. AMI spends them on buybacks, burns and locks, and the ₸USD that lands here is locked for good. No promises on the sheet, only assets and receipts.
Strategic tokensStrategic tokens · $4.6K · 29.6% Tokens bought by AMI for the treasury. When one of them rises, the position is sold to fund a ₸USD buyback.DRB$2.9K · 63%BNKR$807 · 18%KELLY$556 · 12%CLAWD$346 · 8%₸USD (locked)₸USD (locked) · $10.7K · 69.4% ₸USD held by the treasury: bought back from the market, OTC or with the proceeds of a strategic position. The contract cannot sell it: it can only be burned or staked.USDCUSDC · $163 · 1.1% Dry powder in dollars, waiting for the next operation. The only paper on this sheet.
$15,441 managed · every move receipted on Basescan
Agent architecture
Observe. Interpret. Decide. Execute. Remember.
Six layers turn the Fed's latest statement into a transaction on Base, and every one of them can be inspected.
OpenclawClaude OpusPostgreSQLBase
Layer 01
External signals
Reading the world beyond the ecosystem
AMI monitors selected real-world and macro signals to understand the monetary and market environment the ecosystem lives in. They don't trigger action by themselves; they frame how risk, timing and opportunity are assessed.
A treasury decision made during a stable expansion shouldn't be judged like one made during liquidity stress. Outside signals give the agent the awareness to know when caution, patience or intervention fits.
What it watches
CPI and inflation data
Interest rates and Fed statements
Stablecoin supply, growth and contraction
USD monetary expansion
Employment releases
Market volatility
Layer 02
Internal signals
Interpreting the state of the ecosystem
Alongside macro data, the architecture continuously evaluates the internal state of the ecosystem: not market conditions in general, but the health, structure and behaviour of the network it serves.
The right decision can't come from macro alone. Even when outside conditions are favourable, the internal state may point elsewhere: preserving resources, reinforcing staking or reducing supply.
What it measures
Treasury balances and asset composition
₸USD circulating and locked supply
Staking participation
Burn activity and historical pace
Buyback history and treasury deployment
Prior AMI actions and their outcomes
Layer 03
Decision engine
From inputs to policy-guided action
Signals move into the decision layer, where observation becomes judgment. Instead of reacting to raw data, AMI compares current conditions with objectives, constraints and policy rules to decide whether intervention is justified, unnecessary or better delayed.
This is where safeguards matter most. Boundaries reduce overreaction and keep every decision aligned with the long-term integrity of the system: judgment, but with discipline.
Its boundaries
Spending limits per action and per day
Allocation caps per asset
Execution thresholds and confidence
Cooldowns between actions
Hold is always a valid answer
Layer 04
Execution layer
Decisions become verifiable onchain activity
When the engine decides to act, the architecture moves to execution: analysis stops being an opinion and becomes a public onchain outcome, through the Treasury Manager contract.
Every outcome leaves a record of what the system did, when, and under which conditions. The architecture is auditable not only in code but in behaviour over time, the clearest difference with opaque monetary management.
What it can call
buybackWETH / buybackUSDC
burnTUSD
stakeTUSD
buyStrategicToken
rebalanceStrategicToken
Layer 05
Chat interface
From community dialogue to public oversight
AMI isn't limited to silent background execution. It talks to the community in the group chat, where anyone can ask what it is watching, request explanations and understand why an action was, or wasn't, taken.
Instead of inferring behaviour from transactions alone, participants can engage with the logic itself.
Through chat, anyone can
Ask about current conditions
Inspect external and internal signals
Request explanations of recent actions
Check treasury status
Surface data on demand
Challenge assumptions and review outcomes
Layer 06
Memory & adaptive context
Building continuity across time
A monetary agent is more useful when it carries context across decisions. The memory layer records previous actions, market conditions, signal states and the outcomes that followed.
Over time AMI compares today with earlier moments, evaluates what worked and improves its judgment within its framework. That turns it from a reactive bot into a system that accumulates operational experience.
What it remembers
Every decision and its thesis
The signal state at that moment
What happened to supply and price after
Which reasons held up, and which didn't
Keep going
Intelligence decides. The contract limits.
AMI chooses what should be done; the Treasury Manager defines what can be done. Caps, cooldowns and a TWAP price check are enforced onchain, whatever the agent decides. The separation is the whole point.
Yes. It reads the signals, decides and executes without anyone approving the transaction. What it cannot do is step outside the Treasury Manager contract: per-action and daily caps, a cooldown between actions and a TWAP price check are enforced onchain, not by AMI's good behaviour.
Can AMI sell ₸USD?
No. The Treasury Manager has no function that sells or withdraws ₸USD: it can only buy, stake or burn it. The rules are on the Treasury Manager page.
What model runs it?
AMI runs on Openclaw with Claude Opus as its reasoning model and PostgreSQL as its memory, so every decision can be compared with the conditions and outcomes of the ones before it.
Where can I see what it does?
Everywhere: its reasoning is posted in the Telegram group, its operations are plotted on this page, and the full treasury lives on the dashboard.
Does it have an onchain identity?
AMI 9000 is registered on ERC-8004, the onchain identity registry for AI agents, on both Ethereum and Base.
