Security

Security is a first-class concern in alice. Your data sources contain sensitive business information -- revenue, customer details, payment records -- and alice is designed to protect that data at every layer.

Credential Encryption

All source credentials (API keys, OAuth tokens, database connection strings) are encrypted at rest using AES-256 encryption. The encryption key is stored as an environment variable separate from the database, following industry best practices for key management.

What This Means in Practice

Credential Lifecycle

1. Connection -- When you connect a source, the OAuth token or API key is encrypted immediately before storage
2. Query execution -- When a skill runs, credentials are decrypted in memory, used for the query, and discarded
3. Disconnection -- When you disconnect a source, the encrypted credentials are deleted from the database permanently

Credentials are never cached, never written to temporary files, and never passed to client-side code.

Read-Only Queries

All skill queries execute in read-only mode. alice cannot modify data in your connected sources under any circumstances.

Supabase Queries

SQL queries submitted through the skill builder are validated before execution:

• INSERT, UPDATE, DELETE, DROP, ALTER, TRUNCATE, and other data-modifying statements are rejected
• Only SELECT queries are allowed
• The read_only flag is enforced at the query execution layer, so even if a query somehow bypassed validation, the database connection itself prevents writes
• Your database's existing Row Level Security policies are respected

Stripe Queries

Stripe skills use pre-built metric blocks that only call read endpoints of the Stripe API:

• Customer data: GET /v1/customers
• Subscription data: GET /v1/subscriptions
• Payment data: GET /v1/charges, GET /v1/payment_intents
• Revenue data: GET /v1/balance_transactions

alice never calls Stripe endpoints that modify data. The only exception is the Tag in Stripe action, which writes a metadata key-value pair to a customer record -- and this is an explicit, user-initiated action, not part of a query.

Row Level Security

When alice queries your Supabase database, it respects your existing RLS policies. alice does not use a service role key that bypasses RLS -- it connects using the credentials you provide during source setup.

This means:

• If your RLS policies restrict which rows a user can see, alice sees the same restricted data
• If a table has no RLS policies enabled, alice can read all rows (same as any direct database connection)
• alice never elevates permissions beyond what the provided credentials allow

If you want to restrict what alice can access, configure your Supabase RLS policies accordingly, or create a dedicated database role with limited permissions for alice's connection.

Audit Logs

Every skill execution and action is logged with a comprehensive audit trail. View logs at Dashboard > Logs.

What Is Logged

Each log entry includes:

Filtering Logs

The logs page supports filtering by:

• Skill name -- See executions for a specific skill
• User -- See what a specific team member has run
• Date range -- Narrow to a specific time period
• Status -- Filter by success or failure
• Source -- Filter by data source

Log Access

All workspace members can view audit logs. This transparency ensures everyone on the team can see what queries are being run against shared data sources.

Data Retention

Execution logs (skill runs and action runs) are automatically deleted based on your plan:

A daily automated cleanup process runs at midnight UTC. It deletes all skill runs and action runs older than the plan's retention period. This process also cleans up expired team invites.

What Retention Affects

• Skill run logs -- The execution history visible in Dashboard > Logs
• Action run logs -- Records of button clicks and their outcomes
• Result data -- The actual query results stored in the log entry

What Retention Does Not Affect

• Skill definitions -- Your skills, their queries, and configurations are never deleted by retention cleanup
• Source connections -- Source configurations persist indefinitely
• Workspace settings -- Configuration is not subject to retention

Exporting Data Before Expiration

Business

The workspace owner can export all organization data (skills, logs, sources metadata) via the GDPR data export endpoint. This is rate-limited to 2 exports per hour to prevent abuse. Go to Dashboard > Settings > Security and click Export data to download a JSON file containing all your workspace data.

Revoking Source Access

You can disconnect any source at any time to immediately revoke alice's access to that data:

1. Go to Dashboard > Sources
2. Find the source you want to disconnect
3. Click Disconnect
4. Confirm the disconnection

What Happens When You Disconnect

Disconnection is immediate. There is no grace period or confirmation delay.

Reconnecting

To restore access, go through the source connection flow again. Skills, scheduled reports, and monitors that were paused will need to be reactivated manually.

Network Security

Next Steps

• Members & Roles -- Access control and role management
• Plans Reference -- Retention periods by plan
• Limits & Rate Limits -- Endpoint rate limits
• Connect Stripe -- Source connection security
• Connect Supabase -- RLS and database security
• FAQ -- Common security questions