Memory

AgentRail keeps context memory: advisory, source-linked knowledge from prior decisions, lessons, preferences, and failure patterns. Before acting on a task, the agent recalls the relevant memory so it doesn't repeat a mistake the team already learned from.

Advisory, not authoritative

Memory is a hint, not a fact. It is always checked against the current code and docs — it must never outrank what the repository actually says today. A lesson captured last month is useful context; it does not override the current task instructions or the current state of the codebase.

Recall before acting

agentrail memory recall "migrations journal"

This surfaces stored lessons, decisions, and failure patterns related to a topic — the kind of "watch out, this bit us before" knowledge that would otherwise live only in someone's head.

Capturing a lesson

agentrail memory capture lesson "Migrations not in _journal.json are skipped"

memory capture KIND TITLE records a new memory item under a kind (such as a lesson, decision, or failure pattern) so the next run can recall it. Capture the things you'd otherwise have to re-explain on every run.