Policies

Policies define conditions that trigger automated responses, from alerts to blocking an action entirely, and they run server-side on every signed action for real-time control over what your agents can do. Use the built-in engine or bring your own rules. Either way Asqav signs the outcome into a tamper-evident receipt, recorded by a party that is not the agent's operator.

Creating Policies

Create a policy with a name, condition, and action:

bash

# Create a rate limit policy
curl -X POST https://api.asqav.com/api/v1/policies \
  -H "X-API-Key: sk_live_your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "api-rate-limit",
    "description": "Limit API calls to 100 per minute",
    "action_pattern": "api:*",
    "conditions": {"rate_limit": "100/minute"},
    "action": "block_and_alert",
    "severity": "high"
  }'

# Response includes the policy id and active flag
{"id": "pol_abc123", "name": "api-rate-limit", "is_active": true, ...}

Policy Conditions

Every policy matches actions by action_pattern, a glob matched against the action_type (for example api:* or data:delete:*; * matches everything). The optional conditions object narrows the trigger further:

Condition key	Value Format	Description
`rate_limit`	`{count}/{period}`	Triggers when an agent exceeds the action rate. Periods: `minute`, `hour`, `day`
`payload_contains`	Plain text string	Triggers when action metadata contains the specified string
`payload_regex`	Regular expression	Triggers when action metadata matches the regex pattern
`time_window`	`HH:MM-HH:MM`	Triggers when actions occur outside the allowed time window (UTC)

payload_contains and payload_regex inspect the action context, so they cannot evaluate on hash-only signs: the payload never reaches the API. Rate limits and time windows still evaluate on every sign. When a hash-only sign matches a policy that carries payload conditions, the receipt's controls_evaluated.policy entry records the limitation with payload_conditions_not_evaluated: true. See Control Attestation.

For action_pattern, the SDK resolves semantic names like sql-destructive or file-write to their glob equivalents with asqav.resolve_pattern("sql-destructive") (returns data:delete:*). See Enforcement for the full list of built-in patterns.

bash

# Block destructive SQL (glob from asqav.resolve_pattern("sql-destructive"))
curl -X POST https://api.asqav.com/api/v1/policies \
  -H "X-API-Key: sk_live_your_key" \
  -H "Content-Type: application/json" \
  -d '{"name": "block-destructive-sql", "action_pattern": "data:delete:*", "action": "block"}'

# Block actions containing sensitive keywords
curl -X POST https://api.asqav.com/api/v1/policies \
  -H "X-API-Key: sk_live_your_key" \
  -H "Content-Type: application/json" \
  -d '{"name": "block-delete-operations", "conditions": {"payload_contains": "DROP TABLE"}, "action": "block"}'

# Regex match for credit card numbers
curl -X POST https://api.asqav.com/api/v1/policies \
  -H "X-API-Key: sk_live_your_key" \
  -H "Content-Type: application/json" \
  -d '{"name": "detect-credit-cards", "conditions": {"payload_regex": "\\b\\d{4}[- ]?\\d{4}[- ]?\\d{4}[- ]?\\d{4}\\b"}, "action": "alert"}'

Policy Actions

Actions define what happens when a condition is triggered:

Action	Behavior	Agent Impact
`alert`	Logs an alert, action proceeds	None - agent continues normally
`block`	Rejects the action silently	Action fails, agent remains active
`block_and_alert`	Rejects the action and logs an alert	Action fails, alert created
`suspend_agent`	Suspends the agent immediately	All future actions rejected until unsuspended

Listing Policies

bash

# List all policies
curl https://api.asqav.com/api/v1/policies \
  -H "X-API-Key: sk_live_your_key"

# Response (truncated)
[
  {
    "id": "pol_abc123",
    "name": "api-rate-limit",
    "action_pattern": "api:*",
    "conditions": {"rate_limit": "100/minute"},
    "action": "block_and_alert",
    "is_active": true,
    "violation_count": 0
  }
]

Toggling Policies

Enable or disable policies without deleting them:

bash

# Toggle a policy between active and inactive
curl -X POST https://api.asqav.com/api/v1/policies/pol_abc123/toggle \
  -H "X-API-Key: sk_live_your_key"

# Check current state
curl https://api.asqav.com/api/v1/policies/pol_abc123 \
  -H "X-API-Key: sk_live_your_key"

Tip

Start with the alert action to monitor policy triggers before switching to block. This lets you tune conditions without disrupting agent operations.

Auto-Remediation

Agents are automatically suspended after exceeding a violation threshold. Suspension runs on all plans. Automatic quarantine on critical alerts is available on Enterprise.

You can configure the suspension threshold and time window:

bash

# Get current config
curl https://api.asqav.com/api/v1/remediation/config

# Update threshold
curl -X PUT https://api.asqav.com/api/v1/remediation/config \
  -d '{"auto_suspend_threshold": 10, "auto_suspend_window_seconds": 7200}'

Setting	Default	Range	Description
`auto_suspend_threshold`	5	1-100	Violations before auto-suspend
`auto_suspend_window_seconds`	3600	60-86400	Rolling window for counting violations

Per-action save policy

The action_type_policies table lets a regulated org set retention shape per action_type. A wire-transfer action can carry the full payload for seven years while support-ticket replies are stored hash-only for ninety days, in the same organization, without a separate cloud install.

The default behaviour, when no row matches a given action_type, is the existing org-level retention mode. The cloud applies the policy at sign time, regardless of what the SDK requested.

save_mode	Effect at sign time
`full`	Raw payload and metadata persisted, subject to the org retention window.
`hash`	Raw payload dropped at sign time. Only canonical bytes hash and the signature persist.

sql

INSERT INTO action_type_policies (organization_id, action_type, save_mode, retention_seconds)
VALUES ('org_abc123', 'support.ticket.reply', 'hash', 7776000);

Manage policies from the terminal with asqav policies (list, create, delete).