What changes
Organisational memory remains available over time.
Explicit entities, relationships, provenance and history can be queried again when an authorised application or agent needs them.
Platform capability · Roadmap
Organisational intelligence that stays inside your boundary.
Sovereign Intelligence is the planned intelligence layer across SovereignEU applications: permission-aware retrieval from a time-aware organisational graph, customer-selected models and policy-controlled tools working from relevant, source-linked evidence.
Status: Roadmap. Sovereign Intelligence is a planned shared capability, not a currently available universal memory or autonomous decision system. Supported sources, retrieval controls, models, tools and operating modes will be published with applicable releases.
Persistent context, finite prompts
Context that outlives the prompt.
Language models still have finite input limits. SovereignEU is designed to preserve approved knowledge and history in a temporal graph, then assemble a task-specific evidence set for each request.
What does not change
The model still receives a bounded prompt.
A temporal graph does not make a model context window infinite. It retrieves a relevant, permission-filtered evidence subgraph for the user, purpose and task.
A graph is not a neural network. Its entities, relationships, versions and provenance are explicit, queryable and reviewable rather than hidden in learned model parameters.
A governed context flow
From approved knowledge to reviewable evidence.
Each stage narrows authority and retains a path back to the source. A shared data foundation does not grant universal access.
Connect
Represent approved records, work, policies, versions and relationships over time without erasing application ownership.
Retrieve
Select a bounded, permission-checked subgraph through graph, text, vector and time-aware queries.
Reason
Let specialised agents investigate defined questions and return findings linked to permitted sources.
Act
Use authorised tools only within purpose, policy, approval and operational boundaries.
Evidence
Retain permitted citations, actions, approvals, versions and outcomes for review.
Multi-agent reasoning
Many agents. One governed evidence trail.
Specialised agents can examine different graph neighbourhoods and time periods in parallel. Their findings can be combined with provenance and stated uncertainty; access controls remain enforced during retrieval and tool execution.
Explore AssistantsPermission boundary
Shared context is not universal access.
Each application remains responsible for its records. Purpose, identity, permissions, retention and legal constraints still apply. Context is assembled for a specific user and task, not copied into an unrestricted global memory.
Explore SovereignEU DBCustomer-operated intelligence
Keep the complete path inside the selected boundary.
Storage, retrieval, inference, orchestration, tool execution and evidence remain local only when every configured dependency and operational path does the same.
Operational controls
Context remains governed from retrieval to action.
The selected deployment must document what is connected, who can retrieve it, which models and tools may receive it and what evidence is retained.
Purpose and identity
Resolve the user, agent, purpose and current organisational role before retrieval.
Source permissions
Enforce the application’s access rules while selecting graph evidence.
Provenance and time
Return authoritative sources, applicable versions and relationship history.
Model routing
Send bounded evidence only to an approved inference endpoint.
Tool authority
Limit connectors and consequential actions through policy and approval.
Retention and exit
Define logs, citations, deletion, export, backup and handover responsibilities.
A practical first step
Define the context your AI is allowed to use.
Start with one task, its approved evidence, its permission boundary and the operators accountable for the result.