Persistent AI memory, governed at the enterprise
Engram Enterprise turns conversations into structured organisational memory across six categories – facts, entities, decisions, events, insights, and preferences. The result is an AI environment that becomes more useful over time, while keeping every memory visible, governed, and revocable.
The dirty secret of enterprise AI rollouts
Daily active usage of internal LLM tools collapses after the first month. The reason is rarely the model and almost never the interface. It’s that every conversation starts from zero. The model that helped you draft an investment memo on Tuesday has no idea who you are on Wednesday.
People stop using tools that don’t remember them. The first three queries are interesting. The fourth time you have to explain that you work on the credit-risk team, you stop opening the tab.
Engram Enterprise gives the model a memory of each user – what they work on, who they work with, what they’ve already asked, and what context they’ve already provided. The third conversation picks up where the second one left off.
Six memory types. One governed record per team.
Facts
Stable pieces of information that hold true across time. “The London office uses MiFID II reporting templates v3.2.” “Acme’s fiscal year ends 31 March.”
Entities
People, accounts, products, projects, jurisdictions, and the relationships between them. “Jane Doe leads the credit-risk team.” “The Acme matter is under English law.”
Decisions
Conclusions reached and the reasoning behind them. “Approved Vendor X for the security review on 12 March.” “Rejected the merger structure under option B because of Solvency II treatment.”
Events
Things that happened – meetings, filings, releases, milestones. “Q3 board meeting on 14 October.” “ESG report submitted to auditor on 28 February.”
Insights
Patterns and conclusions drawn across conversations. “Healthcare clients ask about PHI handling in week one – prepare the BAA early.”
Preferences
How a person or team likes to work. “Compliance team prefers conformity reports in PDF, not Word.” “Use British English in client-facing drafts.”
Memory that earns its keep – and shows its work
Engram is built around four maturity stages. New deployments start with capture, then progress as the organisation’s memory base grows.
Semantic Recall Lab. A purpose-built tool for compliance and IT teams to inspect, test, and tune retrieval before it goes to end users. Every retrieved memory is shown with its source conversation, confidence score, and access path. No black-box memory.
Capturing
The system extracts memory candidates from conversations across all six types. Users review and confirm what gets retained.
Organising
Memories cluster by entity, project, and team. Duplicates are merged. Conflicts are flagged for human decision.
Recalling
The Semantic Recall Lab gives administrators a sandbox to test what the system retrieves for any given query, with confidence scores exposed and citations linked back to the originating conversation.
Compounding
Recall accuracy rises as the corpus grows. The model knows more about how the organisation works than it did the month before, without anyone having to write it down.
Sphere AI Guide 2026
The practical handbook for deploying AI inside regulated enterprises.
Memory and privacy aren’t a trade-off. They’re a design choice.
The first question every CISO asks about persistent memory is “where is this data and who can see it?” Engram is engineered around the answer.
User and team scoping
Memories can be private to a user, shared within a team, or marked organisational – controlled by explicit access rules, never by accident.
Permission-aware recall
Engram never surfaces memories tied to documents or projects the user has lost access to. Permission changes propagate in real time.
Connectors with native context
Engram ingests structured context from SharePoint, Jira, Gmail, Google Calendar, and Outlook, applying the source system’s access rules at retrieval time.
GDPR-native
Article 17 (right to erasure), Article 20 (right to data portability), and Article 25 (data protection by design) are first-class capabilities, not bolt-ons.
Audit-logged
Every memory creation, retrieval, edit, and deletion is logged to the same immutable audit log the rest of SphereIQ uses.
Where structured memory changes the workflow
An associate at a law firm asks for the indemnity language used in last quarter’s credit agreements. Engram retrieves the decision (which clause was approved), the entities (which client, which jurisdiction), the events (when each agreement closed), and the preferences (this partner prefers narrower indemnities). Four memory types, one query, no re-uploading.
A relationship manager at an EU asset manager opens the tool Monday morning. The system surfaces the insights from last week’s portfolio reviews, the events scheduled in the next ten days, and the facts about each client’s mandate structure. The week starts already in context.
A compliance officer at a hospital network asks which clinical AI systems still need an Annex III classification. Engram returns the entities (each AI system), the decisions (which have been classified), the events (when the last review happened), and the preferences (the CCO wants outstanding items in PDF, not email). The handoff to next quarter’s audit takes minutes.
Sphere in Numbers
We understand that actions speak louder than words and numbers but here are some key facts about us.
Get the Right Talent now0
Years of Excellence
0+
Projects Delivered
0
Countries
Globally diverse, community-focused
0+
Clients
top 20 average 8+ years
We'd love to hear from you!
Please provide your contact details, and our team will get back to you promptly.