Olbrain - The Agent OS your enterprise can trust

1. Executive Summary

Olbrain is the Agent OS for enterprises: the platform on which enterprise agent systems are built, run and governed. An enterprise enters through one business objective and can expand onto the same operating layer as further systems prove useful.

Our differentiation is trust, built on self-identity. A model supplies reasoning; it is not the continuing agent. An enterprise must be able to change a model without losing the actor to which a mandate, operating history and actions belong. Olbrain's CNE architecture defines that self-identity through Coherence, Narrative Continuity and Exclusivity. Permissions govern what the agent may do; an identity-bound record supports attribution of what it did. The enterprise remains accountable. The signed record is live; full cryptographic enforcement of continuity and exclusivity remains in development, as detailed in Section 8.

The growth strategy connects a repeatable entry with depth inside the account. We are testing which buyer, recurring problem, initial agent system and delivery method repeatedly convert into paid use and expansion. Reusable product capabilities and deployment knowledge should then help that pattern travel across enterprises and countries.

Build from India. Compete globally. India is our initial engineering and relationship base. Selected international validation can overlap with domestic commercial proof; broader deployment capacity scales with demonstrated demand, delivery economics and financing.

The long-term destination is The Machine Brain: the enterprise's own cognition, made coherent. Olbrain's architectural thesis is that progressively connecting governed systems, continuing agent identities and appropriately shared context can give an enterprise a more coherent way to understand its state and coordinate action. This is a direction of development, not a shipped enterprise-wide capability or evidence of enterprise-wide learning. Agent OS is the operating layer; Alchemist directs specialist agents on it; scoped agent systems perform the business work. FP&A is the initial commercial path toward that broader ambition.

Agent system — a single-agent or multi-agent application built for one defined business objective, together with the data, tools, workflows and governance required to operate it.

Enterprise adoption provides category context. Deloitte describes the transition toward coordinated enterprise agent systems, Gartner highlights cost, value and control barriers, and Wonderful's September 2026 financing illustrates investor interest in enterprise AI operating platforms. These sources support the context; Olbrain's market scope and operating plan are defined separately in Section 6. [1] [2] [3]

2. Problem Statement

Making an agent useful once is different from operating an agent system repeatedly inside an enterprise. The work crosses systems, data, permissions and human decisions. Each additional workflow introduces operating and governance requirements that must be maintained as the business and its models change.

Continuity becomes an operating requirement as roles change. Enterprises already know, remember, decide and learn through people, processes and systems. Our thesis is that replacing people in some roles with disposable agent processes can weaken continuity of mandate, context and attribution. Each system therefore needs an identifiable actor and agreed controls: who acted, under what authority, what informed a decision and how accepted corrections carry forward. The enterprise retains accountability; human judgment and authorization remain where the business requires them. This is Olbrain's reasoning about the problem, not proof that every enterprise requires the same architecture.

We use an autonomy ladder to describe this progression. It is a framework for assessing work and control requirements, not a forecast that every workflow should become fully autonomous. Bounded tasks can deliver value; greater autonomy is appropriate only where the outcome and oversight justify it.

Olbrain's illustrative L0–L5 autonomy ladder is separate from Boulding's taxonomy of system complexity discussed in our essay; the numbers do not map onto each other. Deloitte's enterprise blueprint provides context for the transition from bounded tasks to coordinated systems; the previous penetration percentages are omitted because their exact attribution has not been verified. [1]

The six-agent FP&A system in our current engagements tests multistep coordination. Other recorded systems predominantly address bounded tasks. Functional capability, production deployment and paid adoption remain distinct stages.

Without reusable operating foundations, integration and control work must be repeated across deployments. Gartner forecasts that more than 40% of agentic AI projects will be cancelled by the end of 2027, citing cost, unclear value or insufficient risk controls. This is category context, not a forecast of Olbrain's outcomes. [2]

3. Solution

Olbrain builds, runs and governs enterprise agent systems on one operating layer. The enterprise starts with a defined business objective. Integrations, operating history and reusable capabilities established for that system provide a foundation for subsequent systems; declining delivery effort is an outcome to measure.

Build. Alchemist, the super-agent that operates the platform, directs specialist agents for discovery, building, support and observability. Humans provide business context, direct the work, review results and approve the system.

Run. Published agents run on Olbrain's cloud with state, memory, hosting, observability and lifecycle support. The enterprise determines its objectives, permissions and approval responsibilities.

Screens. Every agent system has an API; a human workspace is applied where people need to review, decide or act. Screens and systems have a many-to-many relationship. P&L and HR screens are live and reusable across customers. Catalog lineage and fork-and-modify are roadmap capabilities, distinguished in Section 8.

Clear roles on one OS. Studio is where agent systems are built; Noesis runs and governs them. Alchemist is the super-agent that directs specialist agents to operate the OS, rather than a separate enterprise brain. Agent systems carry out scoped business objectives. Screens add a human workspace when needed; headless systems do not require Screens. The Machine Brain names the long-term ambition for the enterprise's cognition as these systems are progressively integrated, not another product to install.

Platform interaction modes and current implementation offer

• Self-serve: the customer's own people direct the build agents and operate the resulting systems.
• Forward-deployed: Olbrain FDEs direct the initial build and deployment, then transfer the operating pattern.
• Co-build: customer teams and Olbrain FDEs work together.

These describe modes of using one product, not three implementation packages currently offered. The current enterprise offer is implementation by Olbrain forward-deployed engineers (FDEs) alongside customer finance and IT teams; customer participation and handover are agreed within that scope. Building and testing are free within qualified, agreed engagement scope; subscription begins at production publish. Observed conversion and support economics are tracked separately for each method.

Govern. Self-identity connects the agent's declared objective, operating history and attributable actions. The identity-bound, signed, append-only record is live. Cryptographic enforcement of exclusivity and narrative continuity remains in development. Access controls determine permissions; self-identity concerns the actor to which a mandate and record belong.

Four needs the identity architecture is designed to address

• The risk owner: identify the actor and mandate associated with an action, and apply the institution's control and oversight processes.
• The incident investigator: check signed records to support attribution. Absence of a receipt alone does not establish that an action could not have occurred; that stronger conclusion requires verified completeness and enforcement across action paths.
• The security team: follow an identity's history through a model or workflow change, with engineering evaluation and revalidation before deployment.
• The customer: continue a relationship with the same agent, carrying appropriate context and commitments across interactions. The commercial benefit beyond ordinary memory must be demonstrated in the customer's workflow.

These needs extend across regulated operations, continuing customer relationships and other recurring enterprise work. They explain the intended value of the architecture; the implementation status in Section 8 defines what is available today. Enterprises retain accountability and human approval responsibilities.

4. Customer Traction

The entire team is concentrated on FP&A at a listed manufacturer, our strongest prospect for a first paying enterprise deployment. Other development work and active follow-ups are paused: the opportunities remain open, but we are no longer driving their testing, change-request and customisation cycles. Urgent inbound requests may be assessed individually. SME / FDE-built remains retired from active acquisition.

Status record updated from the founder update of 14 October 2026. These are company-reported engagement stages, not independently audited results. Publication, customer approval, subscription, invoicing and collected payment are separate events. Requested or scoped expansion is not contracted revenue.

What the manufacturer example establishes

The enterprise builds its FP&A agent system on Olbrain, the Agent OS, to prepare, explain and explore forecasts with traceable assumptions and evidence. Finance retains control of assumptions, approvals and judgment. Forecast build and validation are completed: the system reproduced all 41 P&L lines with 100% accuracy against the agreed benchmark, and the accounts head has cleared this stage. This establishes benchmark reproduction, not predictive accuracy against future actuals. Real-time questions and scenarios are the next capability; management deployment and commercial approval remain pending.

How the FP&A agent system works

Six agents support one forecast process. Five support extraction of inputs, missing-information follow-ups, assumption review, movement investigation and narrative preparation. Cerebellum handles forecast questions and scenario exploration. An orchestrator coordinates the work; a deterministic financial engine computes the P&L from human-confirmed inputs and approved formulas.

Agents propose; finance decides. Only accepted changes apply. Inputs carry source references and calculation methods, and each run is an immutable snapshot. Finance can inspect the evidence behind an assumption, compare movements and explore a scenario without rewriting an approved run.

A management presentation has been prepared alongside the validated forecast, with source traceability. The management review is still ahead. Reporting derives from a selected run using templates, mapping and human review; keeping the forecast and reporting tied to the same run helps finance keep the numbers and explanations consistent for management decisions.

This system combines reusable Olbrain capabilities: specialised agents, deterministic tools, source-linked inputs and workflows with human decisions. FP&A is the entry example; the same Agent OS supports other enterprise objectives.

Requested next steps and longer-term roadmap

Immediate next stage: enable management to question the forecast and explore scenarios in real time, targeted for the week of 19 October 2026. The management presentation and final deployment decision follow; commercial approval remains pending. SAP HANA actuals integration is a requested subsequent step. Forecast accuracy must be evaluated against actual results; improvement is a goal to test, not an established result. Further agents are planned for 24 business drivers, including sales, inventory and raw materials.

Longer term: a continuously refreshed, driver-based rolling outlook for scenario exploration and decision support. Rolling means extending the forecast horizon as time advances. Refresh speed depends on reliable data and the required approvals. A live outlook would remain distinct from approved snapshots used for reporting.

No time saving has been measured. A consistently scoped baseline and comparison are still needed before reporting an efficiency result.

What customer-led building establishes

A research-led consulting firm built and tested its first agent without Olbrain engineers directing the build. This is an early signal of product usability, not evidence of repeatable self-serve deployment. Continued use, paid conversion, support load and production margins remain separate tests.

What customer-requested expansion establishes

The consultancy identified eight more agents after testing its first; the NBFC identified five additions beyond its initial five. These requests show interest in deeper deployment. Payment, operational acceptance and repeated expansion are the commercial evidence still to be earned.

What we stopped pursuing

SME / FDE-built engagements consumed substantial delivery effort relative to small prospective deal sizes and remain retired from active acquisition. Existing commitments and urgent inbound requests are assessed within current capacity. An existing English-and-Telugu voice agent remains a capability demonstration.

Customer-acquisition fallback: high-consideration Shopify merchants

Our already-built AI shopping assistant runs on the same Olbrain platform and helps shoppers gain confidence in choosing the right product. It remains Plan B for customer acquisition if large-enterprise adoption takes longer than expected. A healthcare merchant is targeting 25 October 2026 to go live; this is a customer target, not a completed deployment or established paid use.

Working market estimates are approximately 5 lakh high-consideration Shopify stores globally, including roughly 75,000 with sufficient sales volume to support our pricing. Both estimates remain to be validated.

Other probes span CPaaS, insurance and financial services, healthcare compliance and coworking. They inform the selection decision; their presence does not establish additional validated markets.

5. Commercial Strategy and Value-Risk Proof

Repeat the entry. Expand inside the enterprise.

Olbrain's platform is horizontal. Our commercial entry will concentrate on a repeatable customer pattern: a defined buyer, a recurring business problem, an initial agent system, a value demonstration and a clear expansion path.

Within each enterprise, the first useful system creates a foundation for the next. Across enterprises, reusable product capabilities and deployment knowledge should reduce the effort required to deliver the same class of outcome. These two forms of compounding underpin the growth strategy. Customer data and customer-specific workflows remain isolated; only appropriately reusable patterns and playbooks transfer.

Why the systems belong on one OS. Within an enterprise, shared operating foundations, appropriately scoped context, continuing identities and attributable actions are intended to let each system contribute to the enterprise as a whole. FP&A opens a recurring, cross-department decision process; subsequent scopes can address applicable sales, workforce and resource planning, procurement or inventory drivers before further functions. This path is industry-neutral, including enterprises such as IT services firms; these examples are not mandatory modules or connector-readiness claims. Better coordination, lower delivery effort and paid expansion are hypotheses to test, not benefits already established by preserving identities and records.

The initial segment and entry system are selected together. A broad label such as regulated finance does not identify the particular buyer, job, budget or purchase gate. Our probes must establish those details before the acquisition motion is repeatable.

One segment in focus; two paused

We have concentrated the company on large enterprises, entering through FP&A. We are targeting 2,000 large enterprises in India that undertake financial forecasting. The immediate priority is to complete real-time questions and scenarios at the anchor manufacturer, secure the management decision and establish a paying enterprise deployment. Our aim is to reuse the core agent system while adapting it to each customer's data and processes, then expand into additional agent systems on the same OS. Repeatability and revenue within accounts remain to be demonstrated. Other opportunities are paused; the already-built Shopify shopping assistant is our customer-acquisition fallback.

Architectural fit and commercial pull are assessed separately. Regulated finance provides a strong hypothesis for identity and record requirements, while its current NBFC probe is on hold. Specialist mid-market has an early customer-led build, while repeat use, realised pricing and margins remain unproven. The methods above describe current probes; each segment may use self-serve, co-build or forward-deployment as its needs warrant.

The RBI's June 2026 model-risk draft describes institutional accountability, model inventories, validation and human oversight relevant to this design. It does not endorse Olbrain or establish that our architecture is mandatory. Each deployment must be assessed against the institution's requirements. [5]

Selection rule and evidence gates

The beachhead decision remains December 2026: confirm large-enterprise FP&A, or reopen the paused segments. Selection considers collected recurring payments, realised price, sustained use, paid expansion, sales time, human delivery and support effort, reference strength, and international portability. A first payment alone does not establish replicability. Where the evidence is incomplete, the hypothesis remains open.

The pre-seed targets value proof: customers pay repeatedly at understood prices for systems they continue to use. The institutional seed's primary task is to establish the repeatable customer pattern across similar accounts and develop the acquisition and delivery engine around it. Each round has one primary risk to resolve; product development, selling, retention learning and international validation can overlap.

International validation can begin alongside domestic proof

India is the initial engineering and relationship base; the enterprise market is global. As a repeatable pattern emerges, we intend to test it with selected overseas customers or partners. An international operation of an existing customer, a direct design partner or a suitable delivery partner can provide a bounded portability test. No overseas reference or channel is claimed as established today.

The portable core is the OS, its identity architecture, reusable capabilities and deployment knowledge. Integrations, language, data requirements, procurement, support and regulatory obligations must be qualified for each deployment. Broader international teams and market commitments scale with demand, delivery economics and financing; expansion does not require saturating India first.

Before expanding a route to market, record the buyer and workflow being repeated, what can be reused, the local work required, the way customers will be reached, expected costs, the evidence gate and the capital needed. The pace is set by repeatability and execution capacity.

6. Market Scope and Revenue Scenarios

Market scope follows the work the product can serve. Olbrain's global ambition is distinct from its first beachhead and from the revenue assumed in a three-year plan. All market values must use a stated geography, year, customer definition and annual spending basis; overlapping workflows or buyers must not be counted twice.

The SAM qualification test

Our working focus is recurring enterprise work where continuity of mandate, commitments and attributable history matters. This includes potential regulated operational workflows and continuing customer relationships. Self-identity is a differentiation hypothesis within that demand; a customer must still have a supported job, a buyer and budget, an acceptable deployment path and a measurable reason to purchase.

Qualifying an account requires the customer profile and job to be defined, required integrations and deployment conditions to be supportable, an acquisition and support route to be credible, and expected annual spend to follow an actual commercial model. SAM is not restricted to India by incorporation, and is not automatically every business able to buy software online. Planned capability or country expansion must identify the resources required before entering the serviceable set.

Sizing remains open. Earlier $400B+ enterprise-AI and $50–100B agentic estimates have not been adopted as Olbrain's TAM because their source and category boundaries need reconciliation. Broader AI or labour-market estimates are context, not a substitute for this calculation. A final TAM/SAM number must be supported by the source register and a documented scope.

March 2030 target: the large-enterprise planning case

The financial model uses a single planning case: large enterprises where a recurring FP&A problem opens broader agent-system deployment. Paying accounts rise to 1, 5, 24 and 74 at successive 31 March year-ends, and each active account adds five paying Agents a month, up to 180. At 31 March 2030, 74 accounts and 4,165 paying Agents × ₹19,800 per Agent-month × 12 give ₹98.96 Cr subscription ARR (~₹99 Cr). Metered usage, knowledge-base indexing and storage, organisation platform fees, orchestrator subscriptions and FDE services are excluded from subscription ARR. [7]

Recognised revenue includes ₹19,800 per paying, published Agent per month plus a planning proxy of 5,000 completed responses per Agent-month at ₹2 each, with first revenue in October 2026. Actual usage is counted and billed per published Agent System, not per Agent; the proxy sizes aggregate usage before the system mix is known. Realised usage may differ and is excluded from subscription ARR.

What the planning case assumes

The ~2,000-account universe of large enterprises with recurring FP&A decisions is a working estimate, not a verified serviceable market; account-by-account qualification remains to be completed. Reaching 74 paying accounts requires 4, 19 and 50 net new accounts in successive financial years. A mature account reaches 60, 120 and 180 paying agents at 12, 24 and 36 months after first paid publication; newer accounts keep the portfolio average at 56.3 agents per account in March 2030. The acquisition pace, the publication cadence, customer demand for that depth and delivery capacity all require validation.

A lower-adoption or slower-expansion case must vary customer cohorts, conversion, time to publish, realised price, usage and retention together with costs and financing. A revenue haircut alone does not test all of those dependencies. The first evidence to obtain is paid expansion inside actual accounts. If the December 2026 decision reopens a paused segment, the model is rebuilt for that segment's price, depth and delivery method.

7. Pricing and Delivery Economics

Agent and usage billing is tied to publication, not deployment. Once an organisation is approved, its team, Olbrain FDEs or third-party FDEs may build and test any number of Agent Systems before publication without Agent, orchestrator, response or workflow-step charges. The organisation platform fee applies from approval, knowledge-base indexing and storage is charged from upload while indexed data is retained, and Olbrain FDE time is separately chargeable when requested. Publication starts one 30-day free period for that Agent System; paid Agent, orchestrator, response and workflow-step billing begins on day 31 unless the system is cancelled.

• Organisation platform fee: ₹10,000 per month, prepaid, beginning when Olbrain approves the organisation.
• Published Agent: ₹19,800 per Agent per month, prepaid, beginning after its Agent System's 30-day published free period.
• Orchestrator: ₹2,500 per month, prepaid, when multiple Agents share one end-user interface; the same 30-day published free period applies.
• Conversational usage: ₹2 per completed end-user response, billed postpaid after the free period. One response is counted once at Agent System level even when an orchestrator and several specialist Agents contribute.
• Workflow usage: ₹1 per executed workflow step, billed postpaid after the free period. Each completed step in a published workflow is counted once.
• Knowledge-base indexing and storage: ₹337 per GB per month from upload, charged while indexed knowledge-base data is retained.
• Olbrain FDE services: ₹9,500 per FDE day when requested and approved in a work order. One FDE day means up to eight working hours from one Olbrain FDE, normally scheduled 11:00 AM-7:00 PM IST, Monday-Friday. Customer and third-party FDEs may build on the approved organisation without an Olbrain FDE charge.
• Inference and taxes: language-model costs are included in Olbrain's prices rather than passed through separately. GST and other applicable taxes are additional.

Qualify the work before committing delivery capacity

The operating standard is to agree the first business objective, customer sponsor, required inputs, scope, acceptance criteria, review date and expected subscription/usage pricing before substantial FDE effort. Free pre-publication Agent and usage metering does not imply free organisation access or an open-ended allocation of Olbrain's delivery team. Additional scope is reviewed against the same value and capacity criteria. These are commercial operating requirements being established, not historical conversion results.

A consistent billable unit

The customer purchases an Agent System for one defined business objective. Each Agent inside it must have one clean, unambiguous Core Objective Function. Subscription is calculated from the number of published Agents, with one orchestrator charge where several Agents share an interface. Separately published Agents do not require an orchestrator.

The published Agent System is the usage-metering unit. Analytics may attribute responses and workflow steps to individual Agents, but billing counts each completed end-user response and each executed workflow step once for the system as a whole. The commercial schedule must make the treatment of versions, replicas, paused systems and partial months explicit; customer-specific schedules and waivers override standard pricing only for their stated scope and period. Account-growth projections count paying published Agents, not every Agent built during exploration.

Measure the economics by delivery method

For self-serve, co-build and forward-deployment, track time and human effort to first paid publish, unpaid build effort, conversion, inference and hosting costs, direct support and delivery costs, and subsequent paid expansion. Customer-led building has demonstrated independence from Olbrain engineers in the room; its production cost to serve and margin remain to be measured.

An illustrative five-Agent System behind one shared interface carries ₹101,500 in monthly Agent and orchestrator subscriptions: 5 × ₹19,800 plus ₹2,500. The organisation platform fee, completed-response usage, workflow-step usage, knowledge-base indexing and storage, any Olbrain FDE services and applicable taxes are additional. This illustration is not booked revenue or measured willingness to pay. Inference and direct operating costs must be included when calculating realised margins.

8. Technology and Capability Status

Olbrain owns its orchestration and operating stack. Leading language models supply reasoning; Alchemist directs specialist agents that discover workflows, build systems, support customers and observe operation. Customers and FDEs provide domain judgment, review and approval.

Model-agnostic architecture. Different model providers can be used, but each agent system is engineered around its model's behaviour. Changing the underlying model requires engineering evaluation and revalidation before deployment; continuity of identity does not automatically preserve an approval.

Capability record · reviewed 18 September 2026

Company-reported product status. This table governs capability claims in the memo and its derivatives. A formal definition or roadmap item is not evidence of implementation or independent verification.

Runtime and operating context

The workflow runtime uses a custom Python/FastAPI orchestrator on Cloud Run, with state persisted to Firestore after each step. More than 37 typed step executors compose document processing, reasoning, integration, data operations, control flow, human gates and outputs. Workflow definitions and runs are bound to agent identities so recorded events can be attributed.

Persistent memory uses the Context Service, with append-only, judge-gated admission and GLOBAL → AGENT → SESSION scopes. The CS-packet carries working context, memory and history into a model call. It contains no CNE binding, is not an identity construct and is not a billing unit. The reported token-reduction measure and its limitations are in Section 12.

The PII pipeline combines a DeBERTa-based detector with Presidio and uses AES-SIV tokenisation with KMS-wrapped keys and per-tenant isolation. Its design objective is to protect sensitive inputs before model calls; evaluation must establish detection and handling coverage for each deployment.

Self-identity: the architecture and its intended benefit

Self-identity addresses continuity of the actor across its mandate, history and actions. Coherence concerns consistency with the agent's own past reasoning and Core Objective Function. Narrative Continuity concerns the same identity over committed states. Exclusivity requires at most one live binding per CoF. These formal properties guide the architecture; the table above states implementation maturity.

The signed record supports verification of recorded actions and their association with an identity. It does not by itself establish that every possible action was captured, that a compromised key could never be used, or that every external claimant is excluded. Those guarantees require separate coverage, key-control and enforcement evidence.

Gravitee's February 2026 survey of more than 900 executives and practitioners reported confirmed or suspected agent-security incidents at 88% of responding organisations, and 21.9% of teams treating agents as independent identity-bearing entities. These are vendor-survey findings about security and identity handling; they do not independently validate CNE or measure demand for Olbrain. [6]

Architectural interpretation: continuing agents and enterprise cognition

Our enterprise-cognition essay uses autopoiesis as a conceptual lens for an agent maintaining its identity, memory and narrative through operation. It motivates the continuing actor we want the architecture to support; it is not an independent validation of CNE or a claim of consciousness. The enterprise-wide hypothesis is that governed systems can progressively contribute to a coherent whole while people retain judgment and accountability. The enterprise remains a social organization; this is not a claim of ascent to Boulding's ninth level.

Evaluation must distinguish an attributable record from an authorized, correct and useful action; retained context from a correction that improves subsequent work; and connected systems from demonstrated enterprise-wide learning. Test the relevant controls and outcomes within an agreed scope. None of these outcomes follows automatically from the formal CNE definition or the existence of logs.

The formal target

The following defines the CNE conditions; it is not a claim that their complete cryptographic enforcement is deployed. Formal statement. Identity is defined per Core Objective Function (CoF, written κ). An agent holds a CNE identity at resolution ε and belief threshold b over an interval when three conditions are simultaneously verifier-checkable:

• Narrative Continuity (N) — consecutive committed states in the agent’s GNF chain stay within the declared resolution: d(u_k−1, u_k) ≤ ε.
• Coherence (C) — at every frame, the recorded belief that this is the same self-identity clears the threshold b.
• Exclusivity (E) — |H(κ, t)| ≤ 1: at most one live binding per CoF at any time, globally, across all substrates.

Why Exclusivity cannot be an internal property. Continuity and Coherence are provably preserved under duplication — a bit-perfect copy satisfies both and claims the same self — so no property internal to an agent can individuate it. Exclusivity is therefore an invariant over the population of live claimants rather than a trait of the agent, and is enforceable only by an external frame: the Global Narrative Frame. That is the formal reason exclusivity must be externally framed rather than self-asserted. Full treatment: Formal Foundations of the CNE-Protocol.

Demonstrate the advantage in the customer's work

The evaluation should follow an authorised action through its identity, mandate, recorded version and signed receipt; inspect the history through a controlled change; and test the relevant oversight controls. For a continuing customer relationship, measure whether the system preserves appropriate commitments and context across interactions. Outcomes must be recorded separately from the formal properties intended to support them.

Internal use and engineering evidence

Olbrain uses its own agents for engineering documentation, production debugging and code review, DevOps, hiring and outbound work. A repository audit on 15 October 2026 supports 70+ repositories, 25,000+ commits and 1.2 million+ lines of source code. These indicate build scope; customer use, reliability and economics provide the commercial evidence.

Audit scope: 71 accessible Olbrain repositories at pinned default-branch commits. Their histories sum to 26,728 commits, including merges. The conservative source-file count excludes tests, dependencies, generated files, documentation, scripts, migrations and configuration; cloc 2.10 reports 1,280,543 code lines after removing comments, blank lines and identical files across repositories. This includes product source, website and internal tools; it does not establish that every counted file is deployed in production. The commit sum may include shared history across repositories.

9. Competitive Context and Defensibility

A shared enterprise operating layer is becoming an established product direction. Deloitte describes a more coordinated enterprise agent architecture. Wonderful's 2 September 2026 announcement positions its platform as an enterprise AI OS and reports a $550M Series C at a $5B valuation, operations in more than 35 markets and 650 employees. These are primary-source category signals, not proof of equivalent architecture or Olbrain's commercial repeatability. [1] [3]

Wonderful also announced its India launch on 8 August 2026, with local deployment teams. India provides Olbrain's initial relationships and engineering base; geographical presence alone is not a defensible advantage. [4]

The alternatives a buyer considers

Buyers can extend an existing enterprise software environment, assemble systems with their own engineers, engage a consultancy or systems integrator, or adopt a dedicated agent platform. The relevant comparison is the buyer's actual work: time and effort to useful deployment, ability to continue operating it, controls and evidence, total cost, and expansion into the next system.

Public product descriptions do not establish an audited capability comparison. Where a peer's identity or record properties are not documented in sources we have reviewed, the status is unverified. We do not infer absence from silence or claim an exclusive category position without a substantiated comparison.

What Olbrain must demonstrate

Olbrain combines agent-directed building, operation and governance with an identity-bound, signed record. The commercial case is to demonstrate why this combination improves a particular customer's deployment and continuing use. Current capability boundaries are in Section 8; collected payment and expansion are separate evidence in Sections 4 and 12.

What can compound

• Reusable product capabilities and integration patterns, with measured reductions in subsequent delivery effort.
• Validated operating and control patterns for a repeatable customer profile.
• Production experience that improves reliability, support and deployment playbooks.
• Paid customer expansion and reference relationships that help acquire comparable accounts.
• A useful screen catalog, with future lineage and sharing capabilities treated as roadmap until adopted.

Track these through reuse, deployment time, support effort, realised margins, retention, paid expansion and reference-led acquisition. Customer information remains isolated. No fixed time-to-copy, permanent moat or production-learning lead is asserted without evidence.

Self-identity is the architectural differentiation we intend to turn into a sustained commercial advantage. The pace of learning from real deployments and taking the repeatable pattern internationally matters alongside continued product development. Systems integrators and other partners may become distribution routes; a partner relationship only enters the plan as established when there is evidence of sourced, delivered or paid work.

10. Team

Composition (current, ~10 people): - 2 co-founders (CEO + CTO) - 8 engineers - Founders own enterprise sales, with a chief mentor from PadUp Ventures engaged weekly for 18+ months.

Headcount plan: 10-intern intake batches every 3 months with ~50% retention filter.

Founders:

Alok Gotam

Founder & CEO

Leads category strategy, product narrative, and enterprise GTM. A 25-year intellectual arc on machine agency — four years of self-study, IIM Ahmedabad MBA (2011). Writes regularly on AGI, agency and identity. Co-founded Hotify (AutoML; acquired by Sonasoft). Founded Olbrain in 2017; led Dilaton Technologies as CEO (autonomous UAVs, 2019–2023); registered Olbrain Labs Pvt Ltd in August 2024.

Nishant Singh

Co-Founder & CTO

IIT Kanpur — architect of the core algorithm behind India’s Aadhaar (UID), identity at 1.3-billion scale, with published fingerprint-recognition research. Category winner, Animal-AI Olympics 2019. Leads engineering and system architecture for the Olbrain platform — including the custom workflow runtime and Agency Protocol implementation. 13+ years’ professional partnership with Alok. Brothers.

How the team uses its own platform: Olbrain runs five internal agents for engineering documentation, production debugging, DevOps, hiring and outbound work. This gives the team direct operating experience and informs product development; measured customer outcomes remain the test of commercial advantage.

11. Funding Details and Evidence Gates

Each round has one primary risk to resolve. Product, value, retention, distribution and execution provide the planning sequence. Product work, customer acquisition, expansion and international learning can proceed together; a future risk is not treated as resolved merely because work on it has begun.

• ₹2.6 Cr angel round, March 2025: funded the core platform and initial enterprise engagements. Independent building and working systems establish core product feasibility. New capabilities, reliability at larger scale and the identity-enforcement frontier remain ongoing product work.
• ₹2.4 Cr pre-seed: targets value proof through collected recurring payment, realised pricing, sustained production use, paid expansion and measured delivery economics. The programme is to agentify one large enterprise over twelve months: establish trust by reproducing the finance team's current forecast and validating the first system; deploy deeper into FP&A and across business workflows; then hand over, so the customer's own team adds and refines further systems on Olbrain. These are proposed goals, not guaranteed outcomes.
• Institutional seed: targets a repeatable customer pattern across comparable accounts, an acquisition and delivery engine, and evidence-led expansion.

Current round

Funding receipt status corrected from the founder's 22 September 2026 update. The ₹3L awaiting transfer is excluded from the ₹52L received total. This is a company-reported funding record, not a bank reconciliation; commitment amounts and cash receipts must be tracked separately.

What this round must demonstrate

• Payment: collected recurring subscriptions at documented realised prices, with discounts and unpaid invoices distinguished.
• Continued use: production usage over successive billing periods and evidence of renewal or continued payment.
• Expansion: a customer-approved addition that becomes paid use; requests and scoped additions are tracked separately.
• Delivery economics: human effort to first payment, serving and support costs, and contribution after direct costs, by build method.
• Concentration: an evidence-backed choice of buyer, recurring job, first system and expansion path, tested for portability.

Planning reviews: December 2026 for beachhead selection and initial payment evidence; August 2027 for deeper pricing, expansion and delivery-economics evidence. These preserve the existing plan's review windows and are targets, not promises of revenue or financing. Fundraising preparation can begin as value evidence emerges; funding is not presumed closed at a fixed date.

Next financing: operating case and accelerated expansion

Existing institutional-seed planning case: ₹25–50 Cr, modelled in FY2027-28 (April 2027 – March 2028), subject to value evidence and investor demand. Its purpose is repeatability, account expansion and the acquisition/deployment capacity to support them.

Accelerated international ambition: a substantially larger round, potentially above $50M (₹500 Cr at the memo's planning conversion), is an ambition to evaluate against demonstrated demand. It is not committed financing or a dependency of the existing forecast. It requires its own customer pipeline, regional delivery plan, hiring pace, use of capital, milestones and downside case.

The amount sought should follow the deployment opportunity supported by evidence. ₹25–50 Cr is the existing operating scenario, not a ceiling on global ambition. International validation can start before that round; larger market commitments require funded capacity.

Previous financing: ₹2.6 Cr from Friends of PadUp and Vinners at ₹50 Cr post-money valuation in March 2025, used to build the platform and establish initial engagements.

12. Business Indicators

Company-reported operational record reviewed 18 September 2026; funding receipt status corrected 22 September 2026; cash position updated from the founder update of 14 October 2026. These figures are not a new bank reconciliation or independent audit. Unknown measures remain unknown. Projections and historical figures both use India's April–March financial year.

Illustrative planning metrics

Exit subscription ARR at 31 March: FY2026-27 ₹0.71 Cr; FY2027-28 ₹5.23 Cr; FY2028-29 ₹26.97 Cr; FY2029-30 ₹98.96 Cr. Recognised revenue: ₹0.31 Cr (October–March only), ₹4.02 Cr, ₹22.26 Cr and ₹90.62 Cr respectively. Paying accounts are 1, 5, 24 and 74, and paying Agents 30, 220, 1,135 and 4,165, with the portfolio average rising from 30 to 56.3 Agents per account. These are the FPS planning case, not achieved results. [7]

Further sizing, unit-economics and accelerated-expansion work must distinguish measured results, model assumptions and targets. No projected margin or usage uplift is treated as established commercial performance.

13. Vision and Global Expansion

Vision 2042 describes the long-term architectural ambition. The dates below are roadmap targets; they do not establish commercial traction or completed technical milestones.

The Machine Brain — long-term ambition; 2027 is an aspirational roadmap horizon, not a delivery commitment. The enterprise's own cognition, made coherent: its existing knowledge, memory, decisions and learning brought into a more integrated whole through progressively governed systems on Agent OS. The ambition is per-enterprise, with customer-specific data and workflows remaining isolated. Shared context must follow agreed permissions; human judgment, authorization and enterprise accountability remain. This is not a separate product purchase or a claim of a shipped enterprise-wide capability.

The starting path is FP&A, followed by separately scoped systems that contribute to the business behind the forecast and then further enterprise functions. The manufacturing design partner illustrates an initial step: forecast build and benchmark validation are complete, including reproduction of all 41 P&L lines against the agreed benchmark. Real-time questions and scenarios remain the next stage; management demonstration, deployment and commercial approval remain pending. A data-supply tier has been discussed as subsequent scope. These stages do not demonstrate enterprise-wide cognition or paid expansion; the dated evidence remains in Section 4.

The 2027 horizon requires defined scope, acceptance criteria and evidence before it can become a committed milestone. Identity continuity and signed records are foundations, not sufficient evidence that an enterprise senses, learns or improves as a whole. The essay on enterprise cognition explains the thesis; Section 8 governs what is available today.

The Cybernetic Brain — target 2032. Extending agency from digital workflows into physical environments through embodied agents.

The Positronic Brain — target 2042. The longer-term research ambition for an analogue brain combining hardware and software for sustained independent operation.

Olbrain is being built for a global enterprise market. India provides the initial engineering and relationship base. The first commercial deployments establish evidence that can support international expansion. Selected overseas customers and partners can test a repeatable pattern while domestic commercial proof develops.

The sequence is to establish a useful entry system, demonstrate paid use and expansion, transfer the same customer pattern to another account or geography, and fund larger deployment capacity where the economics support it. Market expansion does not depend on exhausting India first. Self-identity's potential relevance extends from regulated operations to continuing customer relationships and other recurring work where mandate, commitments and attributable history matter over time.

The next financing should match the opportunity demonstrated by those deployments. Larger global ambition is preserved; the evidence gates and operating capacity determine the pace.

14. Glossary

• Agent OS — the platform on which enterprise agent systems are built, run and governed.
• Agent system — one or more specialised agents, plus data, tools, workflows and governance, organised around one defined business objective.
• Screen — a human workspace applied where a job requires review, decisions or action. Optional, many-to-many with agent systems and reusable across customers.
• Self-identity — the actor's identity across its objective, history and actions. CNE defines its formal properties; implementation status is separately recorded in Section 8. External identity and IAM govern authentication and access.
• Agency — the capacity to act under a mandate, with decisions and actions attributable to an identity and subject to a verifiable record. The enterprise retains accountability.
• Agency Protocol / CNE — the self-identity architecture defined by Coherence, Narrative Continuity and Exclusivity. Full cryptographic enforcement of the latter two remains in development.
• CoF — the individual agent's Core Objective Function, its declared objective. The encompassing business objective belongs to the agent system.
• Control loop — how the agent frames work, consults a model, evaluates a response against its objective and context, commits an action and records the outcome.
• Reasoning — supplied by leading language models. Model changes require engineering evaluation and revalidation.
• CS-packet — compiled working context, including memory and history, passed to a model call. It is neither an identity construct nor a billing unit.
• The Machine Brain — the long-term ambition for the enterprise's own cognition to become coherent through progressively integrated, governed systems on Agent OS. Per-enterprise; not a separate product purchase, a shipped enterprise-wide capability or proof of enterprise-wide learning.
• Studio / Noesis / Screens — Studio builds agent systems; Noesis runs and governs them; Screens provide optional human interaction. Headless systems do not require Screens.
• Autopoiesis — self-production and maintenance of a system's boundary, used in Olbrain's essay as a conceptual lens for continuing agent identity. It is not a claim of consciousness or a demonstrated enterprise-wide learning capability.
• Trust infrastructure — the identity and record mechanisms supporting attributable, verifiable actions and enterprise oversight.
• Alchemist — the super-agent that directs Olbrain's specialist agents to operate Agent OS. Distinct from the scoped business agent systems it helps build and operate, and from The Machine Brain as a long-term enterprise-wide ambition.
• Probe / candidate beach / beachhead / wedge — an engagement tests a hypothesis; a candidate beach is a segment under test; the beachhead is the selected segment; the wedge is its repeatable entry point.
• Published / subscribed / paid — deployed to a runtime / recorded subscription relationship / collected payment. These states are not interchangeable, including during trials.
• Exit subscription ARR — paying, published Agents at a 31 March year-end × ₹19,800 per month × 12, as modelled in the FPS. Excludes metered usage, knowledge-base indexing and storage, organisation platform fees, orchestrator subscriptions and FDE services; distinct from recognised revenue during the year and from contracted revenue.

15. Sources and Canonical Record

Source review: 18 September 2026. External sources below support only the specified claims. Company facts remain company-reported; planning assumptions and roadmap items are identified as such.

1. Deloitte — enterprise architecture context. Agentic enterprise 2028: A blueprint for growth, September 2025. Supports the progression toward coordinated enterprise systems. The exact penetration bands formerly used in the memo were not verified from the reviewed page and are not retained as evidence.
2. Gartner — cancellation forecast. Agentic AI project cancellation forecast, 25 June 2025. Forecasts over 40% cancellation by the end of 2027, citing costs, unclear value or inadequate controls; it is a forecast, not an observed failure rate.
3. Wonderful — financing and footprint. Series C company announcement, 2 September 2026. Supports $550M raised, $5B valuation, more than 35 markets, 650 employees and the company's enterprise AI OS positioning. Does not establish its SAM, Olbrain's market size or an audited architecture comparison.
4. Wonderful — India presence. India launch announcement, 8 August 2026. Supports the announced India launch and local deployment approach.
5. RBI — model-risk draft. Release 2026-2027/528, 24 June 2026, and draft text. Supports institutional accountability (8, 45), inventory and retention (21–24), validation (29–33), change control (38–42), and human oversight (60–63). The cited instrument is a draft, not a compliance certification or endorsement of Olbrain.
6. Gravitee — security survey. State of AI Agent Security 2026 report announcement, 4 February 2026. Vendor survey of over 900 executives and practitioners; supports the confirmed-or-suspected incident and identity-handling figures with the qualifications stated in Section 8.
7. Olbrain — financial assumptions. Financial projections, October 2026 model. Source for the FY2026-27 to FY2029-30 large-enterprise planning case, prices, Agent/account counts and funding case. Its ~2,000-account universe, acquisition pace, publication cadence, usage, system mix and margin assumptions are not independently established market facts.
8. Olbrain — formal identity definition. Formal Foundations of the CNE-Protocol. Source for the CNE conditions and their scope. A formal result does not establish implementation maturity.

Thesis update: 17 October 2026. The revised A Brain for a Body That Already Thinks essay supplies Olbrain's reasoning about organizational continuity, agent identity and coherent enterprise cognition. It is a company-authored interpretation, not independent technical validation, evidence of shipped CNE enforcement or commercial proof. This revision updates the investment thesis and reconciles the vision's customer-stage wording with Section 4; it does not change financial assumptions, funding terms or recorded traction. The effective dates of underlying evidence remain attached to their records.

How this memo governs other documents

Section 4 holds the engagement stages, Section 8 holds capability maturity, Section 11 holds financing status and Section 12 holds measured and reported business indicators. The deck, one-pager and sales materials should preserve those stages and qualifications when reusing a claim. Strategy and target changes originate here; numerical financial-model changes must also be reconciled in the FPS.

Every material update should record its effective date and evidence. Customer records should distinguish built, UAT, published, accepted, subscribed, invoiced and paid; company reporting should distinguish recognised revenue from collected cash. Sources for external numbers must state the population, category, geography and year they measure.

Open verification items: updated customer acceptance and expansion records; current cash and fully loaded delivery economics; account-by-account qualification of the ~2,000 large-enterprise working universe; and a consistently scoped TAM/SAM calculation. Unresolved items are not promoted to established facts in derivative materials.

Conventions: ₹100 = $1 for planning conversions. Figures use Indian financial years (April–March). Projections report subscription ARR at each 31 March year-end. Dates attached to plans are targets, not assurances of revenue, PMF or financing.

Olbrain Labs Pvt Ltd · 825-26, Emaar Emerald Plaza, Sector 65, Gurugram-122102 · CIN U62013HR2024PTC123898 Contact: alok@olbrain.com · olbrain.com