Code review with a memory

The reviewer that stops re-flagging what your team already decided is fine.

error_outline

Without memory

AI reviewers flag the same “issue” your team consciously accepted months ago, every single PR. Human reviewers repeat the same style guidance to every new contributor. The team’s real conventions live in merged-PR comment threads nobody rereads.

check_circle

With Rekall

Review decisions become team memory. The AI reviewer recalls that this pattern was debated and accepted (and why), enforces the conventions that are actually rules, and stops relitigating the settled ones.

How it works

Review decisions are remembered

When a flagged pattern is accepted with rationale, that decision — and its reasoning — lands in the decision graph.

Conventions become rules

Recurring guidance codifies into team rules that are injected into every reviewer session, human or AI.

Reviews get quieter and sharper

The reviewer spends its attention on genuinely new problems instead of re-flagging your architecture choices.

See it in action

Settled means settled

[ai-review] PR #482: flag — raw SQL in route handler?

rekall.recall_decisions("raw SQL in handlers")

→ Decision (Jan 20): allowed for read-only report
  queries — ORM overhead measured at 4x for these
  paths. Re-litigated + upheld Mar 3.
[ai-review] suppressed. 2 new findings posted
  instead (missing index, unbounded LIMIT).