# MemoryOps

> MemoryOps is the operating discipline for governed AI-agent memory: capture, source links, scope, redaction, retrieval tests, conflict review, scorecards, and audit.

MemoryOps turns agent memory from a private recall feature into an operated company capability. The goal is not to remember everything. The goal is to make the memories agents rely on current, permissioned, retrievable, and safe to reuse.

## The short definition

MemoryOps is the practice of operating AI-agent memory like a production system. A team decides what should be remembered, who can use it, whether retrieval works, and which memories need review before agent rollouts expand.

In Ambience, MemoryOps starts when a useful decision, convention, skill, pattern, failure, or reference becomes a typed memory with source, scope, redaction state, and audit history.

## Why MemoryOps exists

A memory layer can make agents faster, but it can also replay stale decisions, leak context across boundaries, or make an old workaround sound authoritative. The problem is not only storage. It is memory quality over time.

MemoryOps gives teams a repeatable loop:

- Capture durable memories from approved sources.
- Redact sensitive text before storage.
- Scope memories to personal, team, project, org, or sensitive access.
- Test whether retrieval returns the right memories and withholds the wrong ones.
- Review stale, conflicting, or over-broad context.
- Preserve audit evidence for reads, writes, changes, work orders, and receipts.

## The Ambience MemoryOps loop

Ambience captures durable memories from approved agent sessions and sources, redacts sensitive text before storage, applies scope, then makes every read and write auditable.

The operating layer is MemoryOps: scorecards, generated retrieval probes, hard-negative checks, structure suggestions, work orders, receipts, and dashboard review queues that show whether the memory plane is getting healthier.

## MemoryOps vs agent memory

Agent memory usually describes persistence: an agent can remember something later.

MemoryOps describes operations: the team can prove the memory is current, scoped, retrievable, and safe to use.

That distinction matters in companies because the biggest failure is not forgetting. It is giving the wrong context to the wrong agent or letting stale context quietly steer future work.

## MemoryOps vs ContextOps

ContextOps covers the whole context supply chain for AI systems: sources, extraction, policy, retrieval, packaging, runtime delivery, evaluation, and repair.

MemoryOps is the governed-memory subsystem inside that broader discipline. Ambience starts from MemoryOps because reusable company context is born in decisions, conventions, failures, skills, and source-linked work history.

## Why Ambience

Ambience is company context for AI agents. It implements MemoryOps through scoped memories, MemoryOps snapshots, retrieval checks, hard negatives, work orders, receipts, conflict review, source-linked provenance, and admin review queues.

## Related

- [ContextOps](https://ambience.sh/company-context/contextops)
- [Context management for AI agents](https://ambience.sh/company-context/context-management-for-ai-agents)
- [Company Context Maturity Model](https://ambience.sh/company-context/maturity-model)
- [MemoryOps glossary](https://ambience.sh/glossary/memoryops)
- [How organisational context grows](https://ambience.sh/writing/growing-organisational-context)