Skip to main content

Documentation Index

Fetch the complete documentation index at: https://sage-f6b5014e.mintlify.app/llms.txt

Use this file to discover all available pages before exploring further.

Behavioral learning handles the gray areas; policy rules handle the absolutes. Rules let you encode hard constraints that adjust the risk score directly — block a recipient outright, cap a single transfer, or restrict trading hours — regardless of what the vault has learned.

​
Global Limits

Every vault has a set of limits applied on every proposal. Update them from Telegram (update limits) or PATCH /status.
LimitDefaultEffect when exceeded
maxSingleTx5,000+30 risk
maxHourlyVolume10,000+15 risk
maxDailyVolume20,000+20 risk
maxWeeklyVolume50,000+10 risk
maxDailyTxCount20+15 risk
allowedHoursUTC—+20 if outside
allowedDaysUTC—+20 if outside
unknownRecipientActionreviewapprove / review (+40) / block (+70)
riskThresholdApprove40APPROVE below this
riskThresholdBlock70BLOCK at or above this
learningEnabledtrueWhether patterns update after execution

​
Custom Rules

Beyond global limits, you can define named rules evaluated on every proposal. Each rule has a type, conditions, an action, and a priority.
Rule typeWhat it matches
amount_limitAmount above a max (optionally for a specific token)
recipient_blockA specific recipient address
recipient_whitelistA specific trusted recipient address
time_restrictionSpecific hours / days of week (UTC)
token_restrictionA specific token address
velocity_limitReserved for velocity-based conditions
customReserved for bespoke logic
Each rule’s action is approve, review, or block, and it contributes a riskScoreDelta to the score when triggered.

​
Example: block large transfers

{
  "vault": "VAULT_ADDRESS",
  "name": "Block large transfers",
  "ruleType": "amount_limit",
  "conditions": { "maxAmount": "1000" },
  "action": "block",
  "priority": 10
}
Manage rules from Telegram (list rules, create rule, update rule, delete rule) or directly via the Rules API.
Rules and behavioral factors stack. A trusted recipient subtracts risk; a triggered block rule adds 70+. The final score (clamped 0–100) is mapped to APPROVE / REVIEW / BLOCK by your configured thresholds.