Every team has unwritten context — technology preferences, naming conventions, architectural decisions, migration details. Instead of documenting them all upfront, you can teach Mesrai through natural conversation so future reviews are more relevant and repetitive bad suggestions go away.

How it works

When you talk to Mesrai in PR comments, it detects when you’re stating codebase context or a team preference and automatically saves it as a Memory. Memories are then applied as high-priority context in all future code reviews and conversations.

Explicit teaching

Directly tell Mesrai to remember something:

@mesrai remember: this repo mirrors a third-party API, so some payload fields intentionally stay snake_case.

@mesrai please remember: this service follows hexagonal architecture and keeps adapters at the edge.

Implicit learning

State a preference naturally — Mesrai picks it up:

@mesrai this repo mirrors a third-party API, so some payload fields intentionally stay snake_case.

@mesrai in this service, tests usually live next to the implementation files rather than in a central test folder.

@mesrai the billing module is mid-migration, so prefer incremental fixes over broad refactors.

What Mesrai won’t save

Mesrai is selective about what becomes a memory:

• Temporary instructions (“fix this now”, “skip this for today”)
• Questions (“what’s the deadline?”)
• Debugging chatter (“I see an error”)
• Vague statements without actionable information
• Requests scoped to a single PR or task

Memory scopes

Each memory applies at a specific level:

Approval workflow

If you want to review AI-generated memories before they take effect, enable LLM-generated memories approval in settings. Memories will enter a pending state until you approve them.

Managing memories

Go to Code Review Settings → Mesrai Rules → Memories tab to view, edit, or delete memories. You can also create memories manually from the UI.

For details, see Mesrai Rules — Memories.