Data retention policy
Hark synthesizes and stores memories from your Slack data, which are saved indefinitely until the user deletes them. Hark does not retain Slack conversations or logs, event and conversation data is automatically purged from memory within 30 minutes.
Data archiving and removal policy
Hark respects user's right to be forgotten. Users can request data deletion by asking Hark to forget things - account and/or data can be removed altogether by emailing either cancellations@hark.com or privacy@hark.com
Data storage policy
Hark stores the minimum Slack data needed to act as each user's personal assistant:
1. OAuth credentials. Slack user OAuth tokens are stored encrypted at rest (AES-256, AWS KMS-managed keys) in our primary database, are never logged or exposed to clients, and are deleted when the user disconnects the integration or revokes access.
2. Message content for the user's personal memory. When a user's assistant reads Slack messages on their behalf, relevant message content and LLM-derived summaries may be retained in that user's private memory store to power personalization and recall. This data is strictly scoped to the individual user who authorized the connection — it is never shared across users, workspaces, or tenants, and is never used to train models. Users can view and delete stored memories at any time; disconnecting Slack halts all further access, and users may request deletion of all Slack-derived data.
3. Metadata. We store channel/user identifiers and timestamps needed for deduplication and routing.
All data is hosted on AWS in the United States, encrypted in transit (TLS 1.2+) and at rest (AES-256). Employee access follows least-privilege with SSO/MFA, and access to production data is audit-logged. Our security program is monitored continuously via Vanta, with SOC 2 Type II certification in progress. Data is retained only while the user maintains an active account and connection, and is deleted upon account deletion or user request.
Data center location(s)
United States
Data hosting details
Our data is cloud-hosted, access is limited and strictly managed.
App/service has sub-processors
no
App/service uses large language models (LLM)
yes
LLM model(s) used
We use a variety of US-based commercial and internal models including: Claude Haiku/Opus, Gemini, GPT4o/Luna
LLM retention settings
zero retention policy in place
LLM data tenancy policy
Multi-tenant architecture with logical tenant isolation and regional . Customer prompts and completions are not shared across tenants and are not used for model training. Zero data retention is available for enterprise API customers.
LLM data residency policy
LLM inputs and outputs are not retained, requests may be routed to US or EU servers based on detected user locale.