Where the work becomes intelligence.

Genba is a governed AI operating system for family offices, private equity, venture capital, corporate venture, and hedge funds. It is designed for work that must move beyond drafting an answer: Genba plans and completes a multi-step objective while preserving human authority, a complete audit trail, and firm ownership of every Capability created.

Genba can coordinate work across the investment lifecycle: research and thesis development, sourcing and diligence, investment or portfolio operations, fund reporting, investor relations, compliance, and back-office workflows. It is most valuable when the work crosses sources, systems, and teams but still requires accountable human judgment.

Chatbots and copilots help a person complete individual steps; fixed automations repeat a predefined process. Genba accepts the outcome, assembles specialist AI agents, builds what is missing, governs work against live systems, and leaves the organization with reusable workflows, tools, skills, and institutional knowledge.

Through Tōryō, Genba gives the leader one accountable AI Chief of Staff rather than a collection of disconnected agents. It translates the outcome into a plan, composes and supervises the specialist team, carries unresolved obligations forward, and brings only consequential decisions back for human judgment.

Genba’s control gateway, Daimon, keeps credentials outside the AI agent entirely. It interprets the requested action, checks the firm’s human-governed permissions and approval rules, executes through the approved system connection, and returns only the permitted result.

Yes. Genba attributes every instruction, source, model call, token, policy decision, approval, system action, result, and cost to the objective that caused it. The firm can reconstruct what happened, why it happened, who authorised it, what it cost, and what changed.

Genba carries each assignment forward through Kyōzō, its governed institutional canon. More than documents or model outputs, Kyōzō preserves the workflows, components, tools, skills, decisions, relationships, evaluations, and provenance that authorised people and AI agents can safely compose into future work.

Start with a 30-day Experience Seat: one Principal, a dedicated environment running a fully synthesised firm, and nothing of the firm's own connected. The path from there is ordered. Experience Seat first, then Genba One for one Principal in daily operation, then Genba Team when the people who own the operating work need resource-scoped tenancies, then Genba Institution for firm-wide adoption. A sophisticated production Capability can be co-developed with Genba FDEs on top of One or Team.

Scope it to one high-friction workflow and fix the success criteria before anything is built: a named acceptor, a rubric, and the thresholds that end the proof early. The seat then opens on a fully synthesised firm with no pre-built Capabilities. The Principal directs Tōryō through the whole loop to author the firm's first Capability: objective, plan, approvals, drafts, revisions. It is judged by repeatable runs against the agreed rubric, and the month ends on a scheduled go or no-go.

No. The firm's own systems are not connected during an Experience Seat, so there is nothing to grant. The trial runs on a fully synthesised estate where the complete loop executes: writes, approvals, and consequence included. That is what makes zero production writes a property of the design. With nothing connected, the firm has no read-only scope to negotiate and nothing to police. Approved connections come later, on Genba One, through Daimon and behind the same contracts.

Nothing is pre-built. The seat ships with no Capabilities, because the point of the thirty days is that the Principal authors the firm's first one and sees exactly how such a thing gets made. It is shaped by the firm's own templates, tone, and thresholds. On either outcome the firm keeps what the proof produced: the completed output, the Capability, and the complete record, exported in open formats. On a no-go the environment is then deleted.