Basic Memory is a tool that stores AI conversation context as plain Markdown files in a user-owned knowledge graph, accessible via an MCP server across Claude, ChatGPT, Cursor, and other AI tools. For their cloud product, they adopted a per-tenant Tigris bucket architecture: each user gets an isolated bucket with scoped credentials provisioned in under 20 seconds via the Tigris Partner Integration API. Sync is handled by rclone bisync with zero egress costs, globally distributed reads, and strongly consistent headers to prevent stale data. Bucket snapshots power both point-in-time recovery and a file-level diff/merge UI built on CodeMirror. The architecture replaces N per-user servers with a shared compute pool and per-tenant storage, dramatically reducing costs while preserving strict data isolation.
Table of contents
Knowledge that belongs to you One bucket per brain Sync that pays for itself File-level history, for free What Tigris solved that other clouds didn't What's next Sort: