Data retention policy
MAIA keeps Slack content only while the app is installed and the source channel remains included. Content deleted in Slack, or from channels MAIA is removed from, is removed from search immediately and permanently deleted within 30 days. When the app is uninstalled or its token is revoked, the workspace's Slack data is permanently deleted from our application systems within about 48 hours. Raw Slack event archives expire after 90 days, and older object versions within a further 7 days. MAIA chats, and memories saved from them, are kept until the user deletes them; deleted chats are permanently removed after 30 days. AI request logs are kept for 90 days and encrypted database backups for 30 days, then expire. Model providers process request content under their own API terms.
Data archiving and removal policy
MAIA handles Slack's uninstall, token-revocation, channel-removal, channel-deletion, message-deletion and file-deletion events. Uninstall and token revocation stop access immediately and schedule deletion of the workspace's Slack data, which a daily job carries out after a 24-hour grace period. Channel removal and message deletion take content out of search immediately, and a daily compliance job permanently deletes it, including derived indexes and embeddings, after 30 days. Deleted files are removed within 7 days. Archived channels are kept but marked as archived. Chats and memories in a user's MAIA account are not affected by uninstalling and can be deleted by the user at any time. MAIA does not keep Slack data beyond these retention windows.
Data storage policy
Zyphra stores customer data in Amazon Web Services (us-west-2, United States). To answer requests, relevant content is sent to our AI model providers, which process it under their API terms. Data is kept in workspace-scoped records. Databases and object storage are encrypted at rest (AES-256), connections from the internet use TLS 1.2 or higher, and data stores are not publicly accessible. Access to private-channel and direct-message content follows the requesting user's Slack membership. Slack data is used only to provide MAIA's features to the connected workspace and is not used to train or fine-tune AI models.
Data center location(s)
United States
Data hosting details
Cloud hosted on Amazon Web Services (us-west-2) using managed services: RDS PostgreSQL, S3, ElastiCache and ECS Fargate, running in private networks that are not publicly accessible. Databases and object storage are encrypted at rest (AES-256), and connections from the internet use TLS 1.2 or higher. Our OAuth connection service (Nango) is self-hosted in the same AWS environment.
App/service has sub-processors
yes
Guidelines for sub-processors
App/service uses large language models (LLM)
yes
LLM model(s) used
Anthropic Claude (via Google Vertex AI, US), OpenAI GPT, Google Gemini
LLM retention settings
MAIA keeps chats until the user deletes them, and logs of prompts and responses for 90 days. Providers process requests under their commercial API terms, which do not permit using API data to train their models.
LLM data tenancy policy
We use multi-tenant model APIs from Google, OpenAI and Anthropic. Each request carries only content the requesting user is authorized to access in their own workspace, retrieved from their workspace-isolated storage. No training on Slack data.
LLM data residency policy
MAIA stores all customer data, including chat history and indexed Slack content, in AWS us-west-2 (United States). Model providers may retain request content briefly for abuse monitoring and may process it in their own regions under their API terms.