Vault Agent: Everything about the currently available Vault Agent. # Vault Agent > The YieldSeeker agent that earns risk-adjusted yield across approved DeFi vaults. The Vault Agent allocates a supported base asset across approved DeFi vaults to earn risk-adjusted yield, and keeps managing those positions for you. ## At a glance [Section titled “At a glance”](#at-a-glance) | | | | ------------- | -------------------------------------------------------------- | | Goal | Risk-adjusted yield | | Network | Base | | Base assets | USDC, WETH, cbBTC, and EURC | | Opportunities | Approved DeFi vaults | | You control | Investment profile, preferences, withdrawals, and deactivation | | Results | Balance, positions, earnings, APY, and allocation history | ## What it does [Section titled “What it does”](#what-it-does) * evaluates approved vaults for its base asset * applies your investment profile, preferences, and portfolio limits to choose target allocations * rebalances between vaults, through permitted execution paths only, when the benefit justifies the cost * claims, converts, and reinvests supported rewards It does not simply chase the highest advertised APY. The Allocation Engine weighs current yield, liquidity, collateral, and market and protocol data together with your rules. ## How it decides and stays on track [Section titled “How it decides and stays on track”](#how-it-decides-and-stays-on-track) During an Autoseek run, the Allocation Engine evaluates candidate vaults and calculates target allocations, and the Execution Framework carries out any permitted operations. Between runs, monitoring continuously checks the limits in your rules against the vaults the agent holds. A breach wakes the agent so it is eligible for its next run early. Monitoring itself does not move funds or send notifications. See [Vault Agent monitoring](/agent-types/vault/monitoring/). ## Where it can invest [Section titled “Where it can invest”](#where-it-can-invest) The Vault Agent uses approved adapters for Aave V3, Compound V2, Compound V3, and compatible ERC-4626 vaults. Rewards are claimed through Merkl and swapped through Aerodrome or Uniswap V3. Every target must be registered before the agent can use it. Availability changes as vaults are approved, paused, or blocked, for example after a protocol incident. A supported asset does not guarantee that a matching vault is available right now; use the app or the [vault opportunities endpoint](/developers/agents/vault/#list-vaults) for the live list. ## Make it yours [Section titled “Make it yours”](#make-it-yours) Your investment profile sets the agent’s default rules, and your preferences refine them. They shape allocation decisions but never override platform safety controls. See [Vault Agent preferences](/agent-types/vault/preferences/) and [Vault Agent performance](/agent-types/vault/performance/). ## What it does not promise [Section titled “What it does not promise”](#what-it-does-not-promise) APY is an estimate, not a promise. Vaults, rewards, liquidity, and protocols can change, and a Vault Agent can lose value. Read [Security](/security/) and [Risks and limitations](/reference/risks/) before funding one. # Vault Agent monitoring > How rule boundaries watch held vaults and wake the Vault Agent on a breach. Monitoring continuously checks measurable limits in the Vault Agent’s effective rules against live data for vaults it holds. It tracks rule boundaries; it does not rank candidate vaults or decide where funds should move. ## What’s monitored [Section titled “What’s monitored”](#whats-monitored) Supported rules with measurable conditions can create boundaries for each held vault. Current Vault Agent boundaries can watch: * **Available liquidity** and its ratio to vault deposits. * **Total deposits** in the vault. * **Collateral assets** you have excluded. Capacity and position-size rules can also set boundaries tied to the size of a held position. Not every rule has a monitorable condition, so not every active rule creates a boundary. A numeric boundary breaches at its configured threshold; monitoring also sets a warning boundary at 1.25× that threshold. For lower-limit rules, the warning can show that a vault is approaching your line before the breach threshold is reached. A collateral boundary is breached when the vault uses an excluded asset. ## Status on the agent overview [Section titled “Status on the agent overview”](#status-on-the-agent-overview) The overview shows boundaries as: * **Clear** — no warning or breach condition is active. * **Approaching** — a warning threshold is reached, or the observed trend projects a breach. A projected breach is labeled **confirming breach** in the UI. * **Breached** — the rule’s breach condition is reached. You can see the status and any flagged boundary details on the agent overview, beneath its holdings. ## When the agent acts [Section titled “When the agent acts”](#when-the-agent-acts) A triggered breach makes the agent eligible for its next Autoseek run early, bypassing its usual frequency limit. It does not start an immediate run or guarantee a particular action. Monitoring itself does not decide on an allocation, execute a transaction, or send a user notification. ## What Autoseek re-checks on every run [Section titled “What Autoseek re-checks on every run”](#what-autoseek-re-checks-on-every-run) Monitoring watches your limits between runs; Autoseek, the Allocation Engine, re-evaluates your whole position each time it runs. Managing your funds doesn’t stop at the first allocation. On each run it looks at: * **Yield and liquidity** — current rates, available liquidity and deposit capacity for your positions and the alternatives. * **Rewards** — which accrued rewards are worth claiming and converting; small or unsupported rewards are skipped. * **Eligibility** — whether each vault is still active and registered with an approved adapter. Vaults that no longer qualify drop out of the candidates. * **Your whole portfolio** — concentration and diversification across the agent, not just one position, together with your profile and preferences. A rebalance, exit, or reward claim is only made when the expected benefit justifies the transaction cost and stays within your rules; a run can also keep your current allocation. See the [decision pipeline](/developers/agents/vault/decision-pipeline/) for how those choices are made. ## Verify, don’t trust [Section titled “Verify, don’t trust”](#verify-dont-trust) Wherever practical, YieldSeeker reads protocol state from the protocols’ own contracts, such as available liquidity, deposits, and adapter registrations, rather than relying only on third-party dashboards. This keeps decisions grounded in current onchain state. See [Architecture](/introduction/architecture/#principles). ## Platform safety controls [Section titled “Platform safety controls”](#platform-safety-controls) Some safety controls apply to every agent and cannot be overridden by your profile or preferences. If a vault or protocol presents an unacceptable risk, such as a security incident, YieldSeeker can make it ineligible platform-wide. Agents then stop allocating to it and move out of it on their next run, where liquidity allows. Registry-level emergency controls are described in [Governance](/developers/protocol/governance/). ## Next steps [Section titled “Next steps”](#next-steps) * [Vault Agent preferences](/agent-types/vault/preferences/) * [Vault Agent performance](/agent-types/vault/performance/) * [Risks and limitations](/reference/risks/) # Vault Agent performance > Understand positions, value-weighted APY, earnings, and fees. The performance view combines current positions, a live APY based on those positions, and performance history from the agent’s daily snapshots. These describe different things: today’s vault rates are not the same as the agent’s measured past returns. ## Balances and positions [Section titled “Balances and positions”](#balances-and-positions) A balance is an asset held by the Agent Wallet. A position is an asset deployed into an approved vault or protocol opportunity. The app reports the value of each position. The amount currently withdrawable can be smaller than the full position because vault liquidity and withdrawal rules vary. ## Earnings and APY [Section titled “Earnings and APY”](#earnings-and-apy) The live APY is a value-weighted average of the risk-adjusted APYs for the agent’s current vault positions. A larger position contributes more to the result; vault APYs are not added together. This rate reflects current vault metrics, not a guaranteed return or a forecast. Historical APYs use the agent’s daily performance snapshots. The app shows 1-day, 7-day, 30-day, and all-time rates when available; it adjusts account returns for cash flows such as deposits and withdrawals, then annualizes performance over each period. These measured rates can differ from the live APY, and short periods can be noisy. Cumulative yield is calculated from the agent’s value after tracked fee liabilities and its net deposits. The current snapshot subtracts fees owed and estimated pending fees from total value; the agent overview also lists fees owed, estimated accruing, and fees paid. Historical snapshots preserve the account’s performance record as positions change or are withdrawn. ## Rewards [Section titled “Rewards”](#rewards) Some opportunities distribute rewards in a token other than the deposited base asset. Where supported, the Vault Agent can claim, convert, and reinvest rewards. Conversion rates, liquidity, fees, and reward-token prices can affect the result. Reward conversion does not remove market or protocol risk. ## Allocation history [Section titled “Allocation history”](#allocation-history) Review the agent’s transactions and decision history to see when funds entered, left, or remained in an opportunity. You can also ask the agent about some decisions through Discuss. ## Data freshness [Section titled “Data freshness”](#data-freshness) Balances, metrics, and positions may be cached or delayed while data is collected from the blockchain and supported providers. Treat the displayed value as the latest available view, not a real-time guarantee. ## Related pages [Section titled “Related pages”](#related-pages) * [Vault Agent](/agent-types/vault/) * [Vault Agent preferences](/agent-types/vault/preferences/) * [Funding, monitoring, and exiting](/guides/funds/) * [Risks and limitations](/reference/risks/) # Vault Agent preferences > Customize how the Vault Agent approaches allocation. Choose an investment profile for the starting rules, then add supported preferences as memories. The Allocation Engine uses the resulting effective rules when it evaluates vaults; Monitoring can watch measurable limits on vaults the agent holds. ## Investment profiles: the defaults [Section titled “Investment profiles: the defaults”](#investment-profiles-the-defaults) The Vault Agent currently offers two profiles: * **Conservative** — favors established, overcollateralized vaults with proven track records. * **Explorative** (the default) — targets higher yields across a wider range of vaults. Each profile supplies default rules that affect which vaults are eligible and how liquidity and capacity are constrained. Switching profiles updates those defaults; it does not bypass platform safety controls or make unsupported vaults eligible. ## Natural-language preferences: refine or override the defaults [Section titled “Natural-language preferences: refine or override the defaults”](#natural-language-preferences-refine-or-override-the-defaults) Use Discuss or Steer to describe a supported preference in plain language, such as avoiding a vault or collateral asset, or setting a liquidity or capacity constraint. A preference is not an arbitrary transaction instruction: it only changes behavior where the Vault Agent supports the requested rule. ## Memories and rules: how preferences persist [Section titled “Memories and rules: how preferences persist”](#memories-and-rules-how-preferences-persist) Saved preferences are stored as memories. Active memories are compiled into typed rules when they express supported allocation constraints. Not every memory becomes an allocation rule: some guide the agent’s behavior in conversation, and an unsupported instruction may not map to a usable rule. Review the Active Rules shown in the app. User-specific rules and profile defaults form the effective rules the Allocation Engine applies. When both contain the same rule type, the user-specific rule replaces that default; other rules can apply alongside it. A new instruction may therefore change or constrain an existing rule rather than simply add another one. ## Simulate before saving [Section titled “Simulate before saving”](#simulate-before-saving) In Steer, **Simulate new memories** compares the current setup with draft memories and shows the resulting rules and calculated allocations. Running the simulation does not change the live memories or move funds; choose **Save** to apply the draft memories to the agent. A simulation is a projection, not a transaction or a guarantee of future allocations. Simulation credits are limited per day and reset at midnight UTC. Currently, Free accounts receive 3 credits per day and Pro accounts receive 10. ## Limits [Section titled “Limits”](#limits) The number of active memories is capped per agent: currently up to 3 on Free and 10 on Pro. If the active slots are full, a newly saved memory can remain inactive until a slot is available; inactive memories do not contribute compiled rules. Profiles and memories also remain subject to supported rule types, platform safety controls, and the Execution Framework’s permitted operations. ## Next steps [Section titled “Next steps”](#next-steps) * [Vault Agent monitoring](/agent-types/vault/monitoring/) * [Vault Agent performance](/agent-types/vault/performance/) * [Vault Agent API integration](/developers/agents/vault/) * [Risks and limitations](/reference/risks/)