Skip to content

OrgBrain: a governed layer of intelligence under the decision.

Grounded in an institution's own records, it lets a named human make a better, faster, defensible decision.

Product name

How it is built

One converged database, agents under guardrails, access, governance, and the deployment it runs on. Drag to turn, click any layer or component, pull the layers apart, or play the ten-step tour.

Open the diagram full screen

The diagram is a self-contained page: nothing is loaded from outside this site. Its panel lists every layer, component, flow and tour step, so it reads without WebGL as well.

Swap the model

GPT · GPTA frontier language model named in the strategy as one of the commoditised models every institution can reach through a hyperscaler.Gemini · GeminiA frontier language model named in the strategy as one of the commoditised models every institution can reach through a hyperscaler.Claude · ClaudeA frontier language model named in the strategy as one of the commoditised models every institution can reach through a hyperscaler.Llama · LlamaA frontier language model named in the strategy as one of the commoditised models every institution can reach through a hyperscaler.

Any model, swappable without rebuilding. The knowledge layer and the audit trail stay.

If it cannot cite, it does not answer

Illustrative data

Illustrative registry file — Sample Ministry, Case 00-ILL

  • Page 3

    The filing lists three beneficial owners. The third owner matches a name already on the watch list compiled by the unit last quarter.

  • Page 7

    The unit's standing procedure is to escalate a match to a named analyst before any register action.

Every answer cited.

The four human-control modes

Control modeWhat it means
In-the-loopA person approves every output before it has any effect
On-the-loopThe system acts; a person watches and can step in
In-commandA person sets the goals and limits and decides when the system is used at all
Automated-with-auditThe system acts alone on low-risk work; every action is logged and sampled

Domain agents

  • Case triage

    Proposed

    May. Summarise files, surface precedent, order a queue for review

    May not. Rule, recommend an outcome as final, close a case

    AI risk: Very high · Control mode: In-the-loop

  • Transaction analysis

    Proposed

    May. Link entities, score alerts, draft a referral memo

    May not. File a report, freeze an account, sanction

    AI risk: High · Control mode: In-the-loop

  • Procurement review

    Proposed

    May. Flag anomalies across bids and contracts

    May not. Award, disqualify, debar

    AI risk: High · Control mode: In-the-loop

  • Registry verification

    Proposed

    May. Cross-check filings and ownership chains

    May not. Register, strike off, penalise

    AI risk: Medium · Control mode: On-the-loop

  • Supplier traceability

    Proposed

    May. Trace product and supplier records to source

    May not. Certify or decertify a supplier

    AI risk: Medium · Control mode: On-the-loop

  • Repricing and appeals

    Proposed

    May. Compare claims to contract terms, draft an appeal

    May not. Deny a claim or a benefit

    AI risk: High · Control mode: In-the-loop

  • Proposal drafting

    Proposed

    May. Draft from the client's own documents and donor rules

    May not. Submit, commit budget

    AI risk: Low · Control mode: In-command

  • Verified impact reporting

    Proposed

    May. Trace each reported result to its source page and assemble the ledger

    May not. Sign on the institution's behalf

    AI risk: Medium · Control mode: On-the-loop

Sources on this page

  • S-01Sphere Corporate Strategy, September 2026, Sphereas of 2026-09