Knowledge begins close to the work
The person handling an account, a shipment, or an exception may know what a field really means. A source-specific Endpoint Agent helps turn that explanation into a record that can be checked and reused.
BUSINESS NERVOUS SYSTEM
A business that learns together. Knowledge that stays with the business.
Your people understand things your systems do not explain. OpenLock is being built to connect that human understanding with source-specific agents and shared business memory—so a discovery in one part of the business can lead to a better question, a reviewed interpretation, and a more useful answer somewhere else.
Platform under development. This page explains the intended operating model, not a live product demonstration. Integration scope, capabilities, and service commitments are confirmed for each engagement.
THE IDEA
Not one mind. A connected capacity to learn.
A business nervous system is a way to describe how people, systems, agents, and governed memory can keep learning together. It is not a claim that a company is conscious—or that one AI should make its decisions.
Understanding travels in both directions
Endpoint Agents contribute relevant knowledge to Harmonic Agents. Harmonic Agents relate information across authorized sources and return useful questions to the right endpoint and person. Information does not simply disappear into a central database.
People retain judgment and authority
A suggestion is not automatically policy. A shared interpretation needs the right evidence, scope, and review. Teams can retain legitimate differences rather than being forced to agree with a single AI interpretation.
We prefer this framing to “hive mind”: shared understanding should not erase individual judgment, local context, or access boundaries.
BIOMIMICRY, APPLIED CAREFULLY
Borrow the behavior—not the anatomy.
Biomimicry looks to living systems for design inspiration. Here, the useful inspiration is the nervous system’s combination of local signals, central coordination, return pathways, and adaptation. We translate those ideas into software and operating practices, not a literal digital human. [1–3]
Nerve endings → source-local signals
A changed record, an exception, or a person’s correction can start an inquiry. The Endpoint Agent works close to a meaningful source or business domain, rather than treating every API route as a separate agent.
Spinal pathways → connection and routing
Unlock Connect provides the governed connection between business systems, people, and the Harmonic Layer. The analogy is to the spinal cord’s communication and coordination role—not the bones of the spine, and not a rule that every issue must travel to a central “brain.”
Central integration → cross-business context
Harmonic Agents compare relevant, permitted information across sources. They help surface relationships, contradictions, and questions; they do not replace the people authorized to decide what the business means.
Return pathways → better local questions
A discovery is useful when it comes back to the people who can explain it. A return question carries enough permitted context to clarify the issue without exposing information the recipient should not see.
Adaptation → reviewed memory updates
The software learns by maintaining governed knowledge outside model weights. Changes can add a distinction, correct an error, preserve an exception, or retire an interpretation that no longer applies.
Reflexes → bounded, pre-approved routines
A known validation check or duplicate-event suppression need not wait for a new conversation. Unfamiliar meanings and material changes go to review. This is not permission for agents to make unrestricted operational changes.
The biology is inspiration, not proof of product effectiveness. Permissions, audit records, approval rules, and customer ownership are explicit engineering and organizational requirements, not biological equivalents.
THE COMPLETE LEARNING WORKFLOW
Two conversations. One continuing learning process.
Both conversations remain active throughout the life of the business. They are supported by memory, evidence, and accountable review—not by assuming every conversation is true.
LOCAL LEARNING
Business userEndpoint AgentEndpoint Memory
Explanations go to the endpoint.
Focused questions return to the person.
CROSS-SOURCE LEARNING
Endpoint AgentHarmonic AgentHarmonic Memory
Scoped knowledge + evidence go to Harmonic.
Relevant context + useful questions return.
Loop 1 · Business user ↔ Endpoint Agent
The agent retrieves the permitted source context, identifies what is uncertain, and asks a specific question. The person explains a definition, exception, or decision. Endpoint Memory preserves what was said, where it applies, and whether it is a proposal or approved knowledge. Later questions begin with the current approved record, not a blank conversation.
Loop 2 · Endpoint Agent ↔ Harmonic Agents
The endpoint submits a relevant, versioned contribution—not its entire memory. Harmonic Agents check related knowledge across authorized sources, preserve differences, and identify missing context. Useful observations return to the appropriate endpoint. Its user can clarify, challenge, defer, or reject them; the next reviewed update travels back through the same process.
The connecting step · Review and reuse
The appropriate owners decide which interpretations can be published and for what scope. Approved context can support permitted people, reports, models, and application agents. A later correction or meaningful change reopens the affected knowledge—not every record in the business.
“Complete” means the workflow includes capture, reconciliation, return questions, review, reuse, and lifecycle management. It does not mean OpenLock knows everything about the company.
FOLLOW A LEARNING SIGNAL
One word. Two systems. A better business question.
Illustrative scenario: a CRM marks a customer “inactive,” while authorized accounting records show recent invoices for the same verified customer. What does inactive actually mean?
Illustrative workflow · not a live demonstration
Learn locally
Connect and ask back
Use and keep learning
Read all 12 steps
1. Notice
CRM Endpoint Agent
A label raises a question.
CRM customer DEMO-C104 is marked inactive. No approved local definition explains the field. The endpoint retrieves permitted context before asking for help.
2. Explain
Business user ↔ Endpoint Agent
The person supplies the missing meaning.
The source owner explains: “Inactive means no salesperson assigned in this CRM.” The agent asks where that meaning applies and captures the answer without making it company policy.
“Does this describe only the sales assignment field in CRM A?”
3. Approve locally
Authorized CRM owner
Local authority creates a local record.
The CRM owner reviews examples, scope, and the intended use. The local definition is approved; it says nothing about payment or a company-wide activity threshold.
4. Contribute
Endpoint Agent → Harmonic Agents
Share a scoped contribution—not every memory.
Unlock Connect sends the approved record reference, version, evidence pointers, permitted scope, and dependency information to Harmonic for the agreed customer-view purpose.
5. Relate
Harmonic Agents
Two facts describe different dimensions.
For the verified customer, permitted accounting evidence shows invoices in the selected period. Harmonic proposes that sales assignment and recorded invoice activity be represented separately.
6. Ask back
Harmonic Agents → Endpoint Agent
The discovery becomes a useful question.
Harmonic checks whether the CRM endpoint and intended person may see both the question and its supporting context. A useful, permitted question returns; it does not overwrite local knowledge.
Harmonic Agent → Endpoint Agent → business user
“Should the customer view show sales assignment separately from recent invoice activity?”
7. Clarify again
Endpoint Agent ↔ business owners
The business qualifies the interpretation.
The CRM owner confirms the assignment meaning and desired view. The relevant Finance owner confirms what the invoice evidence supports. The proposal retains the selected reporting period and does not invent a universal “active customer” policy.
8. Approve shared
Required business owners and steward
A reviewed distinction becomes reusable.
Required reviewers approve H-CUSTOMER-001 v1 for the authorized customer view. Harmonic publishes references and permission-scoped projections to relevant endpoints; restricted source details are not copied everywhere.
9. Reuse
Authorized downstream consumer
The next permitted answer starts ahead.
An approved customer-view tool receives the permitted distinction: “No salesperson assigned in CRM; invoices recorded in the selected period.” Source records remain unchanged.
10. Observe
Business user → authorized Unlock Connect intake
Observed use can become another proposal.
A user explicitly records that the two-part view helped resolve the original ambiguity. This is a reported observation, not proof of business impact or causation. The external tool’s conversation is not automatically captured.
11. Change
CRM owner and dependency checks
A changed meaning reopens the right work.
In this fictional timeline, the CRM owner later repurposes the field for operational deactivation. The previous cross-source interpretation is marked needs review and blocked for current unqualified use. Historical access still follows current permissions.
12. Revise and continue
Authorized owners ↔ foundation agents
The business remembers the change, not just the old answer.
Reviewers approve the new source meaning and replace the shared interpretation for its effective period. The customer view uses the dedicated assignment field where authorized, or states that assignment is unknown. A new owner inherits the evidence and open questions.
Assumptions behind this synthetic example
- Identity between CRM and accounting is verified for this synthetic customer.
- Relevant users and services may read/process the specific evidence.
- The actual return question is authorized for its recipient.
- Reviewers hold domain and cross-domain authority; no business approver can waive security policy.
- Invoices demonstrate invoicing, not payment or a universal active-customer rule.
The story in brief
Notice and clarify
The CRM Endpoint Agent retrieves existing definitions and finds no approved answer for this field. It asks the business owner. The owner explains: “In this CRM, inactive means no salesperson assigned. It does not mean the customer stopped buying.” The explanation is first captured as a proposal.
Remember the local meaning
An authorized owner reviews the explanation for this CRM field. Endpoint Memory records its meaning, examples, limitations, ownership, effective period, and approval. A local approval does not make it an enterprise-wide rule.
Relate the wider context
The endpoint contributes the approved distinction. Harmonic Agents compare it with permitted accounting evidence for a verified identity. Recent invoices and sales assignment describe different things; neither should silently overwrite the other.
Ask back
Harmonic returns a question through the endpoint: “Should the customer view show sales assignment separately from recent invoice activity?” The owner clarifies the intended use. The relevant Finance owner verifies what the accounting evidence can support. Missing or restricted evidence remains a limitation.
Review, publish, and reuse
Authorized reviewers approve the scoped distinction. Shared memory links the two concepts and their evidence. An authorized consumer can now distinguish “unassigned in CRM” from “invoices recorded in the selected period,” without changing either source system or inventing a universal customer-activity rule.
Revisit when the business changes
Later, the CRM owner changes how the field is used. Dependent interpretations are flagged for review. A replacement definition receives its own version and effective date; old knowledge is no longer presented as current. The next authorized employee can see what changed and why.
Interactive story: 12 steps, including proposal, local approval, contribution, reconciliation, return question, shared approval, reuse, explicit outcome capture, and a later correction.
Assumptions: customer identity is verified; participants and processing services have appropriate access; the actual return question is permitted; reviewers have the necessary authority. Recent invoices are evidence of invoicing, not automatic proof of payment, profitability, or a company-wide “active customer” definition.
This story uses synthetic records. It is not a customer case study or a performance claim.
MEMORY THAT OUTLASTS A CONVERSATION
Remember the meaning—and the conditions that make it useful.
The durable asset is not merely a chat transcript. It is a governed record of what a statement means, what supports it, who can approve it, where it applies, who may use it, and when it needs another look.
Endpoint Memory · meaning close to the source
Source definitions, mappings, known limitations, exceptions, reviewed examples, open questions, and relevant business explanations. Each record remains linked to its source and responsible people. A source-local interpretation stays local unless an authorized wider use is established.
Harmonic Memory · relationships across sources
Reviewed cross-source relationships, shared definitions, legitimate domain variants, conflicts, dependencies, and decision history. It connects knowledge without assuming that similar words have the same meaning—or that every recipient may see every source.
INSPECT THE SYNTHETIC MEMORY
Illustrative view only—not authentication or live permission enforcement. All example records are invented and included in this public page.
Endpoint record · the initial source meaning
- Statement
- In CRM A, inactive means no salesperson assigned. It does not establish purchasing or payment activity.
- Scope
- CRM A / sales_status / approved customer-view use / original effective period.
- Evidence
- Synthetic source examples and the CRM owner’s explanation.
- Authority
- Approved by the CRM business owner for this source—not an enterprise policy.
- Version / validity
- E-CRM-001 v1 · approved and current at this stage of the story; later superseded.
- Permissions
- Read, process and disclose checked separately for each intended use.
- What stays separate
- Source records are unchanged. No operational write authority is granted.
Shared relationship · the reviewed cross-source distinction
- Statement
- For the verified customer, represent sales assignment separately from recorded invoice activity in the selected period.
- Scope
- Authorized customer view, verified identity, original CRM meaning and explicit reporting period.
- Evidence
- E-CRM-001 v1, E-ACCT-001 v1, verified identity and shared review decision.
- Authority
- Required CRM and Finance owners, with cross-domain stewardship.
- Version / validity
- H-CUSTOMER-001 v1 · approved for this scope at story step 8; not a universal active-customer rule.
- Permissions
- Only permitted projections reach endpoints or downstream consumers.
- What stays separate
- An invoice is not proof of payment or profitability.
Later correction · current meaning replaces the earlier inference
- Statement
- The CRM status now means operational deactivation. Use a separate verified assignment field, or state assignment is unknown.
- Scope
- New effective period and the approved customer view; current permissions still govern historical records.
- Evidence
- Changed source contract, owner explanation, reviewed cross-domain replacement.
- Authority
- New local and shared review decisions; an agent does not overwrite the prior rule on its own.
- Version / validity
- E-CRM-001 v2 and H-CUSTOMER-001 v2 current. Prior versions superseded; affected uses held during review.
- Permissions
- Invalidate controlled stale projections and recheck every intended release.
- What stays separate
- Retained historical evidence is not current authority, and retained data is not automatic access.
Restricted example · no context disclosed
No record or derived question is available in this illustrative view. Use an authorized review workflow to resolve access before any context is returned.
What travels with a memory
Statement and type; source and evidence references; domain, entity, and time scope; owner and approver; review status; read, process, and disclosure policy; version and dependencies; freshness, expiry, and supersession information.
Memory files · readable, portable views
Readable files can help people inspect and transfer the knowledge. In the intended production design, controlled, structured records provide authority; files are versioned views or exports of those records. A model cannot rewrite a memory file and thereby create an approved business rule.
Selective retrieval · not everyone gets everything
Before a new inquiry, the system resolves identity, purpose, permissions, scope, and current versions. It retrieves only the relevant permitted knowledge and checks whether it is approved, disputed, stale, or historical. The chosen model reasons with that context, while the durable memory remains outside its weights.
Endpoint and Harmonic Memory are logical scopes; they do not require separate physical databases. Customer-controlled memory supports continuity across model changes, but replacement models and retrieval indexes still require qualification and testing.
PEOPLE ARE PARTICIPANTS, NOT JUST DATA SOURCES
The system should give something useful back.
Participation is central to the design. The business does not finish “teaching the AI” at onboarding. People keep interpreting changing circumstances—and the system should make that contribution focused, visible, and worthwhile.
Ask a question worth answering
Show why the question matters, what is already known, and what remains uncertain. Prioritize useful questions, combine duplicates, and avoid repeatedly asking for an answer the person has already provided.
Make the person’s options real
Explain, correct, qualify, defer, identify another owner, or say “I don’t know.” Silence is not approval. No contribution automatically grants wider access or authority.
Show what was saved and what changed
Return a visible capture receipt and review status. After approval, show the improved interpretation and its intended use. When a contribution is rejected or deferred, preserve the reason within the permitted scope.
Keep ownership through change
Business owners validate meaning; stewards reconcile cross-domain issues; technical owners validate executable changes; security owners govern access. New owners inherit the relevant evidence, open questions, and review obligations—not unquestioned conclusions.
The design invites broad participation with role-appropriate permissions. It does not require surveillance, unrestricted access, or a separate personal agent for every employee.
CONTINUOUS DOES NOT MEAN CONSTANT CHATTER
Always ready to learn. Quiet when nothing useful has changed.
The learning process has no final graduation date, but individual exchanges must have a purpose and an end. A settled definition can be reused until new evidence, a relevant change, or a scheduled review gives a reason to revisit it.
Start a cycle for a reason
A correction, a new source, schema drift, a conflicting interpretation, an outcome, an ownership change, a permission change, or an overdue review can reopen the relevant question.
Stop when the next step is clear
Resolve the question, publish the approved distinction, leave a scoped uncertainty, wait for evidence, or assign an owner. Bound retries and agent-to-agent exchanges. Repetition is not new evidence.
Maintain what is already known
Keep source data movement reliable, review stale knowledge, trace dependencies, and retire knowledge that no longer applies. Healthy ingestion should not stop just because a learning agent is unavailable.
The intended result is less repeated explanation and fewer recurring interpretation errors—not more agent messages or an ever-growing pile of unreviewed memory.
CONTROL IS PART OF THE WORKFLOW
A question can disclose information. So can an answer.
Every exchange needs more than a connected API. The design makes separate decisions about who may read information, which model or service may process it, and where the result may be disclosed.
A proposal is not an approved fact
Record hypotheses, observations, preferences, definitions, and rules distinctly. Required human or policy-based review depends on scope and impact. Model confidence and agreement between agents cannot grant business authority.
Return only permitted context
A question must not hint at restricted payroll, customer, or commercial information. When useful evidence cannot be disclosed, narrow the inquiry safely or route it to an authorized steward. Do not leak the restricted fact through the question itself.
Preserve legitimate disagreement
Two departments may use “available,” “customer,” or “revenue” differently for valid reasons. Record scope and evidence rather than silently choosing the newest, loudest, or most frequently repeated explanation.
Correction is a first-class operation
Trace a changed interpretation into derived relationships, projections, indexes, and served context. Recheck affected uses, supersede outdated versions, and contain unresolved revocation failures. Audit history does not create permission to keep serving old information.
Learning is not operational write access
Approving knowledge does not authorize a CRM edit, a purchase order, a payment, a pipeline deployment, or a policy change. Those actions require their own scope, controls, and authorization.
Customer-specific knowledge stays within its customer boundary and approved processing arrangements. There is no implicit pooling of one customer’s explanations into another customer’s memory.
Explore sovereignty and controlTHE EXISTING OPENLOCK ARCHITECTURE
A behavior across the architecture—not another layer.
Business Nervous System describes how the existing OpenLock design is intended to behave. The architecture retains three business layers and two integration points.
Business layer
Chaos Layer
The authorized systems, records, documents, files, and working context of the business.
↓ Authorized source information
OPENLOCK CORE
Integration point
Unlock Connect · integration point
Reliable source integration, source-specific Endpoint Agents, Endpoint Memory, and the continuing exchanges with business users and the Harmonic Layer.
↕ Scoped contributions and permitted return questions
Business layer
Harmonic Layer
Governed data and knowledge, Harmonic Agents, shared business memory, cross-source reconciliation, and accountable review.
↓ Approved context
Integration point
Skeleton Key · integration point
Governed outbound disclosure of approved context to authorized downstream consumers. No learning feedback returns through Skeleton Key.
↓ Authorized outward disclosure
Business layer
Agentic Agnostic Layer
Customer-selected, qualified models, agents, applications, analytics, and APIs consuming the foundation.
Top-to-bottom order: Chaos Layer → Unlock Connect ↔ Harmonic Layer → Skeleton Key → Agentic Agnostic Layer. The bidirectional learning exchange is between Unlock Connect and the Harmonic Layer, as well as between business users and Endpoint Agents.
FROM AN IDEA TO A USEFUL WORKFLOW
Start small. Prove the loop. Keep stewarding it.
Begin with one consequential question, the sources needed to understand it, and the people authorized to resolve it. Connect reliable data, establish the source meaning, demonstrate the return question, and verify that the next permitted use can apply the approved distinction.
Implementation establishes the path
Define source scope, owners, permissions, initial knowledge, review responsibilities, and acceptance examples. Retain useful existing data infrastructure rather than assuming a replacement is required.
Managed operations keep it usable
Maintain integrations, handle changes and incidents, support business-context review, and make unanswered questions and stale knowledge visible. The specific operating commitments are agreed for each engagement.
Evidence shows whether it helps
Evaluate owner-reviewed answers, repeated interpretation errors, time to verified clarification, useful cross-source discoveries, expert effort, overdue reviews, and denied unauthorized disclosures. These are proposed evaluation measures—not published OpenLock results.
A successful pilot should demonstrate both learning loops, a permitted reuse, a correction, a restricted-information denial, and memory continuity with a qualified replacement model.
PRACTICAL QUESTIONS
Understanding the boundaries.
Is Business Nervous System a new product or a fourth layer?
No. It is the behavioral explanation of OpenLock’s existing learning design. The three business layers remain Chaos Layer, Harmonic Layer, and Agentic Agnostic Layer. Unlock Connect and Skeleton Key are integration points.
Is the Harmonic Agent the company’s brain?
Only as a limited analogy for coordination. The Harmonic Agent is a logical role that may be implemented by multiple workers. It relates permitted knowledge and returns questions; people and explicit governance retain authority. The company is not controlled by one model.
Where do the memory files live?
The design keeps Endpoint Memory and Harmonic Memory in customer-controlled stores outside model weights and hosted chat histories. Readable memory files are governed views or exports; the precise storage and deployment choices are established during implementation.
Does learning mean automatically retraining an AI model?
No. The core learning workflow updates governed external knowledge. Models retrieve and use appropriate versions. Any model training or model change is a separate, explicitly scoped activity.
Does every conversation become permanent knowledge?
No. Capture, retention, review, and publication are distinct. A useful explanation can become a scoped proposal. Irrelevant or restricted content should not be retained by default, and an external chat is not assumed to have entered this workflow.
Does everyone have to answer questions all day?
No. The intended process prioritizes useful questions, reuses settled knowledge, supports deferral, and assigns review to accountable owners. Continuous learning means sustained readiness to respond to change—not constant interruption.
How does feedback return from an application?
No learning feedback returns through Skeleton Key. A person or explicitly supported, separately authorized intake can submit an observation or correction to the relevant business workflow in Unlock Connect. That submission enters the same proposal, permission, and review process.
What happens when a definition becomes wrong?
The relevant knowledge is reviewed, corrected, superseded, or retired. Dependent uses are rechecked. Historical records can be retained where permitted without being presented as current authority.
Is this already a production capability?
OpenLock’s platform is under development. This page describes the intended operating model. Available capabilities, integration fit, deployment, and service commitments are confirmed for each engagement.
START WITH WHAT YOUR PEOPLE KNOW
What does your business keep having to explain?
Bring one recurring exception, one confusing field, or one decision that depends on knowledge scattered across teams. Let’s explore how that understanding could become a governed, reusable part of your business foundation.
Biological inspiration
- The Biomimicry Institute · What is biomimicry?. Design inspiration from living forms, processes, and ecosystems.
- OpenStax · Anatomy and Physiology 2e, 13.2: The Central Nervous System. Brain and spinal cord as the central nervous system; integration and spinal pathways.
- OpenStax · Anatomy and Physiology 2e, 14.3: Motor Responses. Motor pathways and reflex organization; not all responses require conscious deliberation.
These references support biological inspiration only. The software mappings are design interpretations, not evidence of OpenLock product effectiveness.