Corporate memory for AI is the private, permission-aware history of how an organisation works: decisions, documents, correspondence, campaigns, processes, exceptions and outcomes. It is different from public model knowledge. Public LLM training may understand general language and concepts, but it does not know which proposal was approved, which policy version applies, which customer history matters or which internal decision changed the plan.
## In brief AI becomes more useful in enterprises when it can retrieve the right private knowledge safely. Corporate memory connects Domino and Notes history, SharePoint libraries, Office files and structured work records through permission-aware retrieval. The model reasons over approved evidence rather than relying on public training or whatever a user remembers to paste into a prompt.
## What corporate memory includes Corporate memory is not a single database. It is the living record of work across systems. In Domino and Notes estates, it may include NSF applications, workflow documents, correspondence, mail threads, historical decisions, tasks, approvals and operational logs. In Microsoft environments, it may include SharePoint sites, Teams files, Word reports, Excel models, PowerPoint decks and document libraries. The value lies in context. A policy paragraph is useful; knowing whether it is current, who approved it, which project used it and which exceptions were raised is more useful. Corporate memory gives agents the context needed to answer questions, prepare work and support decisions with evidence.
## Why public model knowledge is not enough Public LLM knowledge is broad but external. It can explain concepts, summarise generic best practice and draft language, but it does not know the private history of your organisation unless you provide it at run time. That creates three problems: the model may make plausible but wrong assumptions, users may paste sensitive information into ungoverned tools, and important internal evidence may be missed. Corporate memory solves a different problem. It gives the model authorised access to the organisation’s own evidence. The model still provides reasoning and language capability, but the factual grounding comes from private systems under enterprise rules.
## Permission-aware retrieval is the control layer A useful corporate memory system must respect permissions. If an employee is not allowed to read a customer file, the AI should not reveal it. If a department owns a restricted SharePoint site, retrieval should follow that boundary. If a Domino NSF has ACL controls, agent access should be constrained by those controls. Permission-aware retrieval also helps with trust. Users can see that answers are grounded in sources they are entitled to access. Reviewers can audit which documents informed a response. Sensitive records can remain excluded from general prompts. This is the difference between enterprise memory and a loose search index.
## Domino, Notes and SharePoint together Many organisations have important history split between Domino/Notes and Microsoft 365. Domino may hold long-running workflows and operational records. SharePoint may hold Office files, templates and collaboration outputs. A useful agent should not force a false choice between them. AI Build’s platform pattern connects these stores through governed retrieval and tool calls. Domino as an AI Agent Platform explains how Domino IDs, ACLs, views and documents support agent execution. OfficeMaker supports the Office and SharePoint document intelligence layer, helping agents retrieve relevant sections rather than whole libraries.
## Why memory needs structure, not just vector search Vector search can be useful, but corporate memory should not be only semantic similarity. Enterprises need deterministic filters as well: date ranges, document status, client, project, owner, classification, version, approval state and permission scope. Structured filters narrow the candidate set before a model reasons over content. This improves accuracy and cost. It also reduces the risk that a semantically similar but outdated document influences an answer. For a cost-focused companion piece, read Enterprise AI Token Economics and the OfficeMaker comparison on Python token comparison.
## How agents use corporate memory in practice A sales agent might prepare an account brief from previous proposals, meeting notes and approved case studies. A service agent might retrieve the latest troubleshooting history and policy rules. A governance agent might compare a requested action with the organisation’s AI policy and past exceptions. A marketing agent might use campaign memory, approved messages and prior performance to draft the next iteration. In each case, the model is not being asked to invent corporate knowledge. It is being asked to reason over approved evidence and produce a useful next step.
## Where to go next Corporate memory is one support layer in the wider Enterprise AI Agent Platform. For Domino-heavy estates, start with HCL Domino Assistant. For cross-system agent design, use AI assistant consulting. For consulting firms and SIs packaging this for clients, see the AI Agent Consulting Partner Blueprint.