Feature Request: Support pluggable database backends (MySQL) for team-shared memory

[github-actions[bot]]

Originally posted as issue #2420 by @xingcheng1061
#2420

Before submitting

• I searched existing issues and confirmed this is not a duplicate

---

Is your feature request related to a problem? Please describe.

Our team works on the same codebase, but claude-mem stores observations in a local SQLite file (~/.claude-mem/claude-mem.db) that is private to each machine. This means knowledge discovered by one engineer — how a module works, architecture decisions, recurring bug patterns — is never shared with teammates. Every engineer starts from scratch, and the collective memory never compounds across the team.

Describe the solution you'd like

Support a configurable external database backend (MySQL or PostgreSQL) so that team members can point to the same database instance and share accumulated project knowledge across sessions.

Concretely:

1. A DATABASE_URL environment variable (or config field) accepting a standard connection string:

[xingcheng1061]

Is your feature request related to a problem? Please describe.
Our team works on the same codebase, but claude-mem stores observations in a local SQLite file (~/.claude-mem/claude-mem.db) that is private to each machine. This means knowledge discovered by one engineer — how a module works, architecture decisions, recurring bug patterns — is never shared with teammates. Every engineer starts from scratch, and the collective memory never compounds across the team.
Describe the solution you'd like
Support a configurable external database backend (MySQL or PostgreSQL) so that team members can point to the same database instance and share accumulated project knowledge across sessions.
Concretely:

1. A DATABASE_URL environment variable (or config field) accepting a standard connection string:
   CLAUDE_MEM_DATABASE_URL=mysql://user:pass@host:3306/claude_mem
   CLAUDE_MEM_DATABASE_URL=postgresql://user:pass@host:5432/cl

2. An abstracted database layer with SQLite as the default driver (zero-config experience unchanged for individual users) and MySQL/PostgreSQL as optional drivers.
   An ORM like Drizzle ORM — which already supports bun:sqlite, MySQL, and PostgreSQL under the same API — could make this relatively straightforward to implement.
   Describe alternatives you've considered

• Shared NFS/network mount + CLAUDE_MEM_DATA_DIR: Pointing all team members to the same SQLite file on a shared filesystem works for read-heavy access, but SQLite's write locking makes concurrent multi-user writes unreliable and risks database corruption.
• Forking the project: Possible, but creates ongoing maintenance burden as upstream evolves.

Additional context
From reading the source code, the SQL used is largely standard and should be compatible with MySQL/PostgreSQL. The two areas that would need driver-specific handling are:

• FTS5 full-text search (SQLite-specific) → MATCH AGAINST on MySQL, tsvector on PostgreSQL
• Migration scripts (v4 → v28) → would need driver-aware execution paths
  This feature would make claude-mem viable for team-scale adoption in professional engineering environments. Happy to contribute or help test if there's interest.