Escalate when the threshold reaches 8.
Approved 2025 · status disputedAssembling route context
Loading…
Loriq CORE / Context General Intelligence
Give every AI and agent the intelligence ofyour organisation.
CORE holds one governed picture of what your organisation knows and hands each model or agent exactly the slice it needs. Current, permissioned, receipts attached. Loriq calls this Context General Intelligence.
Real product / private betaNo fabricated UI
01 / the breakthrough
Context General Intelligence is the capacity for every AI system an organisation runs to work from one shared, current and governed understanding of what the organisation knows, aware of each other, with every answer traceable to its sources.
This intelligence sits outside model weights. It changes as the organisation changes, preserves the evidence behind each answer and remains available when the organisation changes models, agents or vendors.
Get the whitepaper: the argument in fullThe trade / stated once
Model intelligence commoditises. Organisational context compounds.
Axis 01 / model intelligence
Rent the model.
Use the strongest suitable model available and change it when the technology moves.
Axis 02 / organisational intelligence
Own the context.
Your meaning, decisions, authority, relationships and history remain with the organisation.
What becomes possible
Your organisation becomes the intelligence layer.
- 01
No model fine-tuning
- 02
One mind for every agent
- 03
Current as the business changes
- 04
Smaller context for each request
- 05
Answers that lead back to evidence
- 06
A customer-controlled data boundary
Question → context → intelligence
The answer starts before the model sees the prompt.
CORE sits between organisational systems and the intelligence layer. Its job is not to answer in place of the model; its job is to make the organisation legible to it.
Pipeline monitorsix stages / one packet
One continuous shot of a request becoming a packet across the six stages; the stage chips beneath the window light as each phase plays. The live question “Which procedure is current for this team?” is typed at Infer and held on screen. Sources fan in at Retrieve, duplicates merge at Consolidate, and authority, time and verification badges land at Reason, where the two sources whose procedures disagree light up with a conflict seam: the orange v2 evidence persists beside the current v3. Everything compresses to budget at Pack, and the packet docks with a permission stamp at the slot labelled “Chosen model or agent”.
Stage 01
Infer
Resolve the request objective, requesting role and relevant organisational domain.
Stage 02
Retrieve
Fan out across symbolic, semantic, structural, lexical and exact-term signals.
Stage 03
Consolidate
Merge candidates, remove duplicates and preserve the route by which each item arrived.
Stage 04
Reason
Read authority, relationships, lineage, effective time and contradiction evidence.
Stage 05
Pack
Fit the smallest useful source-preserving package to the resolved context budget.
Stage 06
Hand off
Attach that package to the organisation’s chosen model, agent or OpenAI-compatible client.
How coordination works
Act. Write back. Inherit.
When a connected tool acts or a person rules, the outcome enters shared context, with source, authority and effective time attached. The next request from any connected tool receives that state. No re-teaching, no drift, no separate memories growing apart.
- 01
An agent acts, a person rules
A decision, ruling or outcome lands anywhere in the fleet.
- 02
Shared context updates
The outcome is recorded once, versioned, attributed, permissioned.
- 03
Every next request inherits
Each connected tool’s next context package carries the current state, receipts attached.
P-03: Ring with three nodes; a pulse travels act → write back → inherit; a version counter ticks each lap. Fallback: complete ring, all nodes lit.
Interactive thesis / conceptual walkthrough
Ask the organisation.
The same organisation produces different context for different questions. Choose one, then inspect how the real CORE assembly stages change what survives.
Current procedureAuthority to decideWhat changedAuthored example / not customer data
01 / RequestAuthored example
Which procedure is current for this team?
Example data, not a customer record or production product screen.
02 / Retrieval signalsRelative, illustrative
SymbolicStrong
SemanticModerate
StructuralStrong
LexicalSupporting
Relevant care-procedure domain inferredPrior and current versions retrievedAuthority + effective time promotedUnrelated guidance excluded
03 / Bounded packet3 included
Current procedure identifiedStage 4 of 6: Prioritise
Current + authoritativeApproved procedure · v3 · 12 Jul 2026
Named as the approved procedure for this team and effective for the date in the request.
Model-ready answer
Use the approved procedure effective 12 July 2026. The earlier version remains attached as a surfaced conflict.
Sources and exclusions remain attachedFull AI control
Every AI your people use. Seen, governed, approved.
CORE is the gateway your organisation’s AI runs through. Leaders see how teams and individuals use AI, audit any interaction back to its evidence, and decide exactly which models and agents each person may use.
- 01
Usage visibility
See how the organisation and each individual actually use AI, in one governed view.
- 02
Interaction audit
Every request and response is attributable, permissioned and traceable to its evidence.
- 03
Approved AI only
Grant each person exactly the models and agents they may use. Unapproved AI has no path to organisational context.
- 04
Memory without resets
Organisational memory that never resets and never goes stale, governed from day one.
C-02: PROPAGATION: one decision node pulses, a wave travels the edges, and a fleet of agent status chips flips to the current position, one by one.
Inside CORE
Five modules. One shared mind.
Assembly, reasoning and write-back are CORE itself; the modules extend it. They are not separate products and do not imply a finished public interface.
- 01 / CORE Proxy
CORE Proxy
Offers an OpenAI-compatible hand-off path so existing clients can receive assembled context without a new model contract.
- 02 / CORE Connect
CORE Connect
Brings authorised organisational sources into a versioned, relationship-aware context system.
- 03 / CORE Trace
CORE Trace
Carries the source, version and inclusion reason needed to reverse an answer back into evidence.
- 04 / CORE Guard
CORE Guard
Constrains eligible context by the requester and deployment boundary before a packet is built.
- 05 / CORE Eval
CORE Eval
Surfaces contradictory evidence, supports authority and temporal resolution, and preserves unresolved states for people.
Interaction 03 / contradiction across time
Do not hide the hardest part.
Step through the evidence CORE uses to resolve a contradiction. If those signals do not settle it, the system preserves uncertainty and hands the decision to a person.
ConflictLineageAuthorityTimeResolutionseam / live
Escalate when the threshold reaches 6.
Effective 12 Jul 2026 · quality leadEvidence state / Conflict
Two sources look current. They disagree.
CORE does not average them or let semantic similarity decide. The conflict becomes explicit evidence.
When the evidence does not settle itEscalate to a binding human ruling.
No generated certainty. The unresolved state and eventual decision both become part of provenance.
Product boundary
Fits the stack without becoming the stack.
CORE stays neutral across source systems, model vendors and agent frameworks. The organisation owns its context; the intelligence layer can change.
Organisational systems authorised sources
CORE one shared mind
Chosen model or agent OpenAI-compatible path
Direct answers
CORE, without the theatre.
Is CORE another AI model?
No. CORE is a context system around the model or agent an organisation already chooses. It assembles a source-preserving context packet, then hands that packet off.
Is CORE just retrieval?
No. Retrieval supplies candidates. CORE also reasons over organisational structure, authority, time and contradiction evidence, ranks what matters and records what was excluded.
Is CORE an orchestrator?
No. Your agents keep their jobs and their frameworks. CORE is the shared mind they work from. What one does or decides is in context for the others. Task scheduling stays wherever it lives today.
Is there a public product UI?
Not yet. CORE is real and in private beta, but Loriq is dogfooding the engine and improving it through carefully scoped work rather than presenting a fabricated interface.
Where can it run?
CORE deploys inside customer-controlled boundaries. The exact topology is agreed per engagement; this page does not claim a certification or deployment configuration that has not been reviewed.
Does CGI require model fine-tuning?
No. Fine-tuning can shape one model, but it also bakes a snapshot of the organisation into that model. CGI stays outside the weights so context can remain current, traceable and reusable when the model changes.
Is CGI agent memory?
No. Agent memory remembers one tool’s interactions. CGI holds shared organisational context that many assistants, workflows and agents can use.
Does every model receive the whole organisation?
No. CGI assembles the smallest eligible context package for the request. Permissions and deployment boundaries apply before the model receives it.
Bring one consequential context problem.
We scope a narrow source slice, define the evidence and deployment boundaries, and learn against a real organisational question before expanding.