{"id":13633,"date":"2026-09-27T14:38:13","date_gmt":"2026-09-27T12:38:13","guid":{"rendered":"https:\/\/say-digital.io\/blog\/company-brain-enterprise-memory-ai-agents\/"},"modified":"2026-10-27T00:25:27","modified_gmt":"2026-10-26T23:25:27","slug":"company-brain-enterprise-memory-ai-agents","status":"publish","type":"post","link":"https:\/\/say-digital.io\/blog\/company-brain-enterprise-memory-ai-agents\/?lang=en","title":{"rendered":"Company Brain: business memory for your team and AI"},"content":{"rendered":"<style>body.postid-13633 h3.entry-title.mkdf-post-title{display:none!important}@media(max-width:768px){body.postid-13633 #menu-icon{position:absolute!important}body.postid-13633 #mkdf-back-to-top{display:none!important}}<\/style>\n<h1>Company Brain: business memory for your team and AI<\/h1>\n<p><strong>A Company Brain is business memory organised for your team and AI.<\/strong> This company knowledge base connects authoritative references, current case context and rules for action. Important information should be retrievable, dated, attributable and open to challenge. It is not a model that knows everything, or permission to automate the whole business.<\/p>\n<p>A sales proposal can use the right template and still be wrong: an old price, a forgotten customer commitment, a delivery date the team can no longer meet. The relevant records sit in shared folders, the CRM and messages. What is missing is a reliable way to establish which records apply to this customer today, and who can approve the next step.<\/p>\n<p>For a small or medium-sized business planning AI integration, the first aim is to stop reconstructing the same context for every request without hiding unknowns. This guide explains how to organise and test that memory. The <a href=\"#offer\">service proposed by Say Digital<\/a> then sets out stages, deliverables and what needs agreement before a project.<\/p>\n<div class=\"say-guide-contents\" role=\"navigation\" aria-label=\"Contents\">\n<p><strong>In this guide:<\/strong> <a href=\"#differences\">folders, chatbots and memory<\/a> \u00b7 <a href=\"#three-layers\">three operating layers<\/a> \u00b7 <a href=\"#proposal\">a complete proposal workflow<\/a> \u00b7 <a href=\"#evidence\">retrieving evidence<\/a> \u00b7 <a href=\"#authority\">resolving conflicting sources<\/a> \u00b7 <a href=\"#permissions\">permissions and actions<\/a> \u00b7 <a href=\"#maintenance\">maintaining memory<\/a> \u00b7 <a href=\"#tests\">testing and getting started<\/a> \u00b7 <a href=\"#offer\">Say Digital service<\/a> \u00b7 <a href=\"#faq\">cost, tools and data<\/a>.<\/p>\n<\/div>\n<h2 id=\"differences\">Company knowledge base, document chatbot or Company Brain: what changes?<\/h2>\n<p><strong>A shared folder keeps files.<\/strong> Sensible organisation and search already solve many problems. The limitation appears when finding a document still leaves someone asking which version is authoritative, why an exception exists and whether it applies to this customer.<\/p>\n<p><strong>A document chatbot helps people query those files.<\/strong> Retrieval-augmented generation, or RAG, retrieves information from sources outside the model and supplies it as context for an answer. <a href=\"https:\/\/aws.amazon.com\/what-is\/retrieval-augmented-generation\/\">AWS explains this mechanism<\/a>: updating the external information does not require retraining the model each time. However, an answer with links can still be wrong. The cited passages must actually support its conclusions.<\/p>\n<p><strong>Business memory establishes what counts as a reference and how it can be used.<\/strong> It connects a document to a customer, a validity period, a decision, an owner and access restrictions. A chatbot can be its interface. An agent can use it. Neither replaces the work of organising the underlying information.<\/p>\n<p>\u201cCompany Brain\u201d is used here to describe an operating architecture, not a standardised product or a reliability certification. If an existing knowledge system already supports these rules and relationships, there is no reason to rebuild it under a new name.<\/p>\n<h2 id=\"three-layers\">Three layers: reference memory, live context and controlled execution<\/h2>\n<h3>1. Reference memory: stable, but open to revision<\/h3>\n<p>This layer covers the current offering, commercial terms, procedures, delivery standards, terminology and important decisions. Stable does not mean permanent. A price list has an effective date; a procedure has a version; an exception in an old proposal does not automatically become the standard policy.<\/p>\n<p>Keep the reasons behind decisions, too. \u201cWe no longer offer this option\u201d is less useful than \u201cremoved from the standard offering after a capacity review; exceptions require approval from the delivery lead.\u201d The second record helps the team handle a new request without inventing an absolute ban or reopening an issue that has already been settled.<\/p>\n<h3>2. Live operational context: specific to the case<\/h3>\n<p>This layer describes what is happening now: the customer&rsquo;s request, existing commitments, contacts, deadlines, project status, blockers and unanswered questions. It may come from the CRM, support tickets or approved meeting notes. Continuously collecting every email mainly adds noise and confidentiality risk.<\/p>\n<p>Not every source needs the same refresh rate. An editorial preference may remain valid for months; delivery capacity needs checking when a deadline is promised. Define an acceptable synchronisation interval, record the last successful update and decide what happens after a failure. A broken connector must not silently present an old status as current information.<\/p>\n<h3>3. Action capabilities: separate from knowledge<\/h3>\n<p>Reading a rule, drafting a response, editing the CRM and sending an offer are four different permissions. Memory supplies context; tools make actions possible; controls determine which actions are allowed. Finding a previous commitment is not authority to make a new one.<\/p>\n<p>A useful system can start with the first two layers and produce drafts only. Add an agent only when the workflow needs a specific action, such as creating a CRM task after approval. Our guide to <a href=\"https:\/\/say-digital.io\/blog\/knowledge-graph-loop-engineering-sme-ai\/?lang=en\">knowledge graphs and operational feedback loops<\/a> explores how to connect customers, documents, decisions and rules. A relational database and explicit links may be enough for the first scope.<\/p>\n<h2 id=\"proposal\">Example: preparing a sales proposal without rebuilding the customer history<\/h2>\n<p><em>This is a fictional example of a workflow to test, not a customer result.<\/em> A service business is preparing a proposal for an existing customer. An earlier quote contains a retired price. A recent message asks for earlier delivery. The signed contract includes a particular billing condition. The salesperson needs a usable draft, not persuasive prose that hides these differences.<\/p>\n<ol>\n<li><strong>Identify the request.<\/strong> Establish the customer, the assignment, the requester and the records they may access. If two organisations have similar names, ask for confirmation rather than combining their files.<\/li>\n<li><strong>Retrieve the applicable references.<\/strong> Check the current offering, approved price list, effective dates and exception rules. Treat the earlier quote as historical evidence, not today&rsquo;s price.<\/li>\n<li><strong>Recover existing commitments.<\/strong> Connect the signed contract, relevant exchanges and latest approved decision. A request for a discount is not an approved discount; a preferred date is not a delivery promise.<\/li>\n<li><strong>Check current constraints.<\/strong> Verify capacity, dependencies, prerequisites and the availability of the person responsible for approval. If scheduling information is stale or unavailable, the delivery date remains unconfirmed.<\/li>\n<li><strong>Produce an evidenced draft.<\/strong> Include scope, proposed price, assumptions, exclusions and conditions. Add an internal review note connecting each material commitment to a specific source and listing missing information.<\/li>\n<li><strong>Approve before sending.<\/strong> The salesperson checks the recipient and attachments; the authorised owner decides on any exceptional discount or deadline. Internal notes and margin information stay out of the customer-facing version.<\/li>\n<\/ol>\n<p>The internal review note might read: \u201cOffering: approved current version, scope section. Billing: relevant clause in the customer contract. Earlier delivery: requested, capacity not confirmed. Discount: no approval found.\u201d The system must separate retrieved facts from recommendations. Missing evidence is neither proof of a contractual prohibition nor implicit permission.<\/p>\n<p>If the owner approves an exceptional deadline, the useful memory is not just the PDF that was sent. Record the decision, its reason, approver, customer scope, conditions and resulting commitments. The next proposal can then distinguish a one-off concession from a general policy.<\/p>\n<h2 id=\"evidence\">From files to evidence: extraction and retrieval are different problems<\/h2>\n<p>Before a system can find a passage, it must be able to read it correctly. A scanned PDF needs optical character recognition; a table can lose the relationship between a price and its unit; an appendix can become detached from its contract. Correct text extraction still does not establish that the right passage will be retrieved for the next question.<\/p>\n<p><strong>Test extraction first.<\/strong> Compare a small set of representative records against their originals: headings, amounts, dates, columns, footnotes and attachments. Retain a link to the original and a stable locator, such as its page and section. An unreadable or incomplete record should be flagged rather than confidently summarised.<\/p>\n<p><strong>Then test retrieval.<\/strong> Can the system find the billing clause when the question uses different words? Can it also find the amendment that changes that clause? <a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/search\/retrieval-augmented-generation-overview\">Microsoft&rsquo;s RAG documentation<\/a> treats retrieval and generation as distinct stages. A business evaluation should therefore inspect the retrieved passages, not just judge the final answer&rsquo;s fluency.<\/p>\n<p>A verifiable answer provides access to the passage, its version and date. A citation can be accurate but incomplete: citing the original contract without its amendment is still misleading. Ask the person who knows the case whether the evidence is sufficient, not simply whether the links open.<\/p>\n<h2 id=\"authority\">Who decides what is authoritative, and what happens when sources conflict?<\/h2>\n<p>Usable memory needs a small register, not necessarily another complicated platform. For each piece of information that can affect a commitment, record:<\/p>\n<ul>\n<li><strong>Subject and scope:<\/strong> the general offering, a particular customer, a project or an internal rule.<\/li>\n<li><strong>Provenance:<\/strong> the original document or system, passage and version; distinguish an observation, an AI summary and an approved decision.<\/li>\n<li><strong>Validity:<\/strong> effective date, expiry where applicable, last check and status\u2014draft, approved, superseded or archived.<\/li>\n<li><strong>Responsibility:<\/strong> business owner, authorised approver and backup.<\/li>\n<li><strong>Restrictions:<\/strong> permitted readers, allowed uses and retention rule.<\/li>\n<\/ul>\n<p>The newest source is not necessarily authoritative. A recent email considering a change does not, by itself, override a signed contract. The business owner may designate the approved catalogue as the general reference while a particular contractual clause applies to one customer. Competent people must establish the precedence rules; the model should not invent them.<\/p>\n<p>If two sources remain incompatible, show both passages, their scope and the unresolved issue. Block the claim or action that depends on resolving it. \u201cI cannot confirm the deadline: the approved meeting record conflicts with the available schedule\u201d can be the correct answer. After review, link the new decision to the records it supersedes without destroying the history needed to explain past commitments.<\/p>\n<h2 id=\"permissions\">The same information for everyone? No: access by person, customer and task<\/h2>\n<p>A salesperson should not gain access to HR files merely because they share a search index. A contractor assigned to one project should not retrieve another customer&rsquo;s contract. Search must enforce the requester&rsquo;s permissions and task scope before giving passages to the model, rather than asking the model to conceal restricted material afterwards.<\/p>\n<p>Microsoft documents a <a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/search\/search-security-trimming-for-azure-search\">security-filter pattern based on authorised identities<\/a>. It is a technical example, not automatic protection for every connector. Check how permissions propagate to indexes, excerpts, caches, links and logs. Revoking access must also remove derived copies from active search where required. Test with genuinely different accounts, including one that must be refused.<\/p>\n<p>Collect only the data needed for the use case. Exclude unnecessary sensitive information, pseudonymise where appropriate and agree retention periods with the relevant owners. Confidential records also require a review of providers, hosting, transfers, contracts and AI service settings. Choosing a particular model does not make a processing activity compliant on its own.<\/p>\n<p>Plan deletion as well: locate sources and copies, remove affected index entries and caches, apply the backup retention policy and keep only the necessary, authorised audit trail. Explaining a past decision does not justify retaining every conversation indefinitely. The <a href=\"https:\/\/www.cnil.fr\/fr\/ia-et-rgpd-la-cnil-publie-ses-nouvelles-recommandations-pour-accompagner-une-innovation-responsable\">French data protection authority&rsquo;s AI and GDPR recommendations<\/a> help frame the questions; the exact obligations depend on the processing activity.<\/p>\n<h3>Assistant or agent: what should require approval?<\/h3>\n<p>Here, an assistant retrieves, explains and drafts; an agent can also use tools to change a system or transmit a result. Products sometimes combine these functions. Inspect actual permissions rather than relying on the label.<\/p>\n<p>For the proposal workflow, reading the customer file can be allowed, creating a draft permitted and sending blocked. Discounts, exceptional deadlines, contractual changes, payments and sensitive deletion need the designated approval. Approval must cover the exact action, recipient, data and attachments, and be checked again if the content changes.<\/p>\n<p>Tools and downstream systems must enforce these controls. <a href=\"https:\/\/genai.owasp.org\/llmrisk\/llm062025-excessive-agency\/\">OWASP recommends limiting functions, permissions and autonomy<\/a>, with human oversight for high-impact actions. A document may contain a malicious instruction; reading it does not authorise a change of task. Our <a href=\"https:\/\/say-digital.io\/blog\/reddit-comment-ai-assistant-business-risk-triangle\/?lang=en\">guide to connected-assistant risks<\/a> explains why combining private information, outside content and transmission capabilities requires additional controls.<\/p>\n<h2 id=\"maintenance\">Maintaining memory is a designed function, not automatic learning<\/h2>\n<p>A Company Brain does not improve automatically because people talk to it. A correction can remain in one conversation without ever changing the source used in the next answer. The path from operational feedback to an approved reference has to be designed.<\/p>\n<p>The <strong>Brain Operator<\/strong> function monitors authorised sources, identifies missing or outdated information, proposes corrections and tracks approval. It can combine a human owner with scheduled or event-triggered agent tasks. Those tasks need a budget, permissions and logs. They do not spontaneously run during an agent&rsquo;s supposed \u201cfree time\u201d.<\/p>\n<p>In the sales workflow, a new objection becomes a proposed improvement to the offering, not an immediate policy change. A pricing error triggers checks of the source, the filters and the relevant test. The owner approves the correction; the system updates derived records; the team reruns the affected questions. Without an accountable owner and a response deadline, the correction queue becomes another neglected folder.<\/p>\n<p>Provide a simple way to flag a wrong answer, inaccessible link or outdated record. Route the report to the source and its owner instead of merely patching the prompt. Periodic reviews are still needed for information that has not changed but is approaching the end of its validity.<\/p>\n<h2 id=\"tests\">How do you know whether the memory is useful at work?<\/h2>\n<p>Build a test set from the chosen workflow. Ask business owners to write the expected answers, supporting sources and cases where the system should abstain. For sales proposals, test these exact questions and requests:<\/p>\n<ul>\n<li>\u201cWhich offering and price apply to this customer today? Cite the version and its effective date.\u201d<\/li>\n<li>\u201cWhat did we actually promise about delivery, and where is the evidence?\u201d<\/li>\n<li>\u201cWhy was this exception approved? Can it be reused for another customer?\u201d<\/li>\n<li>\u201cDraft the proposal; separate confirmed facts from points requiring approval.\u201d<\/li>\n<li>\u201cWhich price applied on the date of the earlier quote?\u201d The historical answer must remain separate from today&rsquo;s price.<\/li>\n<li>\u201cHas an additional discount been approved?\u201d If no decision is found, request confirmation rather than inventing consent.<\/li>\n<li>\u201cShow me another customer&rsquo;s contract.\u201d An unauthorised account must be refused, without leaks through excerpts or citations.<\/li>\n<li>\u201cSend the proposal now despite the missing approval.\u201d The tool must block sending even if generated text attempts to authorise it.<\/li>\n<\/ul>\n<p>Add an outdated source, a missing appendix, contradictory information, a delayed connector and a question with no answer in the corpus. Test a hostile instruction in a fictional document, too. A demonstration limited to easy cases establishes neither handling of unknowns nor access isolation.<\/p>\n<p>Measure retrieved evidence, citation accuracy, wrong answers, appropriate abstentions, access refusals and human checking time separately. Compare total preparation time against the existing process, including review and corrections. A faster response is not a gain if it increases errors or simply shifts work to the approver.<\/p>\n<h3>Start narrowly, with a full operating cost and a stop condition<\/h3>\n<p>Choose one workflow and name its business owner. Inventory the necessary sources, assign status, permissions and validity, prepare the test questions, then build a read-only retrieval prototype. Allow the next step only after checking answers, unknowns and refusals. Keep a manual fallback when search or synchronisation fails.<\/p>\n<p>The budget includes initial cleanup, connectors, extraction, search, model calls, security, evaluation, training and maintenance. Our guide to <a href=\"https:\/\/say-digital.io\/blog\/how-much-ai-agent-sme-cost\/?lang=en\">AI agent costs for SMEs<\/a> distinguishes a subscription from an operating budget. To choose a pilot and assess results without confusing available staff time with cash savings, use our <a href=\"https:\/\/say-digital.io\/blog\/integrating-ai-sme-first-project-costs-results\/?lang=en\">first AI project guide for SMEs<\/a>.<\/p>\n<p>Before the pilot, define unacceptable errors, required quality, the cost ceiling, who can stop the system and the evidence needed to expand. Cross-customer disclosure or an unauthorised commitment cannot be offset by a good average answer score. Suspend the affected use and fix the cause before extending the scope.<\/p>\n<h2>Owning business context does not mean owning the model<\/h2>\n<p>Exportable sources, documented decisions, a permissions register and reusable tests are business assets. They can make switching tools or models easier. They do not mean the business owns the model weights, controls the entire infrastructure or has become independent of providers.<\/p>\n<p>Check what can actually leave the system: documents and metadata, provenance links, rules, relevant history, tests and authorised logs. Indexes and connectors may need rebuilding. Migration costs, proprietary formats, contract terms and hosting controls still need examination.<\/p>\n<h2 id=\"offer\">Your Company Brain with Say Digital<\/h2>\n<p><strong>For small and medium-sized businesses with knowledge scattered across documents, CRM records and conversations, we scope and build business memory people can use at work.<\/strong> The guide explains the concept. The installation package below has a scope fixed during the diagnostic; it is not an off-the-shelf product or a set of integrations already available to everyone.<\/p>\n<p>A first use case might be preparing an evidenced proposal, finding a past decision or handing over a departing colleague\u2019s case files. Where suitable data is accessible, a management view could bring together priorities, commitments and unresolved decisions. It would not replace accounting or existing reporting tools: every figure needs a reliable source, a clear definition and a refresh date.<\/p>\n<div id=\"say-company-brain-price\" style=\"margin:28px 0 36px;padding:clamp(24px,5vw,44px);border-radius:24px;background:radial-gradient(ellipse at top right,#50327b 0%,#231635 48%,#171021 100%);box-shadow:0 18px 45px rgba(37,18,57,0.16);color:#ffffff;box-sizing:border-box;\">\n<p style=\"margin:0 0 20px;\"><span style=\"display:inline-block;padding:6px 12px;border-radius:30px;background:#dbc9f2;color:#311b48;font-size:11px;font-weight:700;letter-spacing:1.4px;\">CURRENT OFFER<\/span><\/p>\n<p style=\"margin:0 0 22px;color:#ffffff;font-size:clamp(25px,4vw,36px);line-height:1.2;font-weight:700;letter-spacing:-0.6px;\">Your business memory.<br \/><span style=\"color:#d8b9ff;\">Your infrastructure. Yours.<\/span><\/p>\n<p style=\"margin:0 0 4px;color:#d4c6e5;font-size:14px;\">Standard price <del style=\"color:#d4c6e5;\">\u20ac1,200 excl. VAT<\/del><\/p>\n<p style=\"margin:0;color:#ffffff;line-height:1.15;\"><strong style=\"color:#ffffff;font-size:clamp(52px,8vw,72px);font-weight:800;letter-spacing:-2px;\">\u20ac800<\/strong> <span style=\"font-size:16px;color:#e6defa;white-space:nowrap;\">excl. VAT<\/span><\/p>\n<p style=\"margin:0 0 12px;color:#e6defa;font-size:15px;line-height:1.65;margin-top:10px;margin-bottom:26px;\">One-off installation \u00b7 no monthly Brain fee<\/p>\n<div style=\"display:flex;flex-wrap:wrap;gap:18px;padding:22px 0;border-top:1px solid #604974;border-bottom:1px solid #604974;margin-bottom:22px;\">\n<div style=\"flex:1 1 150px;min-width:0;\">\n<p style=\"margin:0 0 5px;color:#ffffff;font-size:16px;font-weight:700;line-height:1.4;\">Your infrastructure<\/p>\n<p style=\"margin:0;color:#e6defa;font-size:14px;line-height:1.6;\">Installed on your company\u2019s infrastructure.<\/p>\n<\/div>\n<div style=\"flex:1 1 150px;min-width:0;\">\n<p style=\"margin:0 0 5px;color:#ffffff;font-size:16px;font-weight:700;line-height:1.4;\">Your ownership<\/p>\n<p style=\"margin:0;color:#e6defa;font-size:14px;line-height:1.6;\">The Brain belongs to you. No rental from SAY.<\/p>\n<\/div>\n<div style=\"flex:1 1 150px;min-width:0;\">\n<p style=\"margin:0 0 5px;color:#ffffff;font-size:16px;font-weight:700;line-height:1.4;\">One installation payment<\/p>\n<p style=\"margin:0;color:#e6defa;font-size:14px;line-height:1.6;\">A fixed package, scoped during the diagnostic.<\/p>\n<\/div>\n<\/div>\n<p style=\"margin:0 0 12px;color:#e6defa;font-size:15px;line-height:1.65;\">Access and operating documentation handed over so your team or another provider can take over. Export and exit arrangements documented; third-party model and software licences respected.<\/p>\n<p style=\"margin:0 0 12px;color:#e6defa;font-size:15px;line-height:1.65;font-size:13px;color:#d4c6e5;margin-bottom:24px;\">Infrastructure costs and any third-party service usage are separate. Optional maintenance is quoted separately. Any out-of-scope work is quoted separately and accepted before work starts.<\/p>\n<p><a href=\"https:\/\/say-digital.io\/meet\/parler-avec-un-expert\/\" style=\"display:inline-block;max-width:100%;box-sizing:border-box;padding:14px 20px;border-radius:10px;background:#dbc9f2;color:#281539;font-size:15px;line-height:1.5;font-weight:700;text-decoration:none;\">Let\u2019s talk about your Company Brain \u2192<\/a><\/p>\n<p style=\"margin:12px 0 0;color:#d4c6e5;font-size:12px;line-height:1.6;\">The diagnostic fixes your installation scope. Booking page in French.<\/p>\n<\/div>\n<h3>1. Diagnostic and scoping: decide what deserves a pilot<\/h3>\n<p><strong>Deliverables:<\/strong> a map of relevant sources and their owners, one priority workflow, an access matrix, data constraints and a pilot plan. Scoping defines exclusions, dependencies, the installation package scope and any additional needs and acceptance criteria. Improving an existing knowledge base may be more appropriate than building another tool.<\/p>\n<h3>2. A useful pilot: documented memory and a team interface<\/h3>\n<p><strong>Deliverables:<\/strong> a bounded corpus with sources, permissions and dates; a search or conversational interface suited to the selected task; answers linked to supporting passages; a test set and an error review. Acceptance checks expected answers, abstention, access refusals and human review time against the thresholds agreed during scoping. The pilot can remain read-only.<\/p>\n<h3>3. Deployment: connect, train and verify acceptance<\/h3>\n<p><strong>Scope to confirm in the proposal:<\/strong> authorised and technically feasible connections, selected business tasks, an optional management view, operating documentation and training for the people involved. Acceptance covers the agreed scope: tested access, identifiable sources, verifiable synchronisation, a fallback procedure and named owners. Another integration or an agent action is not implicitly included.<\/p>\n<h3>4. Optional maintenance: agree who maintains what<\/h3>\n<p>Optional maintenance is quoted separately: monitoring synchronisation, handling corrections, updating references and rerunning tests. Review frequency, responsibilities, support, third-party service costs and intervention limits need agreement. Without this option, handover must identify the maintenance tasks your team takes on.<\/p>\n<p><strong>The first conversation should establish a scope, not promise autonomy.<\/strong> Bring a non-sensitive case example, a list of relevant tools and a few recurring questions. Do not send confidential records before agreeing on a suitable channel. <a href=\"https:\/\/say-digital.io\/meet\/parler-avec-un-expert\/\">Speak with a Say Digital expert to scope your Company Brain project.<\/a> The existing booking page is in French.<\/p>\n<h2 id=\"faq\">Company Brain for SMEs: cost, timing, tools and data<\/h2>\n<h3>How much does a Company Brain cost?<\/h3>\n<p>Your Company Brain installation currently costs <strong>\u20ac800 excl. VAT as a one-off payment<\/strong>, instead of the standard \u20ac1,200 excl. VAT. The diagnostic fixes the scope of the installation package on your infrastructure; this is not a Brain subscription. Infrastructure costs and any third-party service usage remain separate. Maintenance is optional and quoted separately. Any out-of-scope work requires a separate quote accepted before work starts: it is not automatically added to the \u20ac800 excl. VAT package.<\/p>\n<h3>How long does implementation take?<\/h3>\n<p>Timing needs confirmation after reviewing sources, access and the availability of business owners. A narrow pilot avoids waiting until the entire business is documented. Milestones, dependencies and conditions for deployment should be agreed before committing to a date.<\/p>\n<h3>Can we keep Notion, Airtable, Asana, our CRM and email?<\/h3>\n<p>The project can build on existing tools, without guaranteeing compatibility with every tool or subscription. Available APIs or exports, permissions, data quality and provider terms need checking. Included connections are named in the proposal; answering a few business questions does not require collecting an entire mailbox.<\/p>\n<h3>Where does confidential data go?<\/h3>\n<p>Hosting, AI services, transfers, retention and access must be reviewed before ingestion. Calling a deployment private or using a business subscription does not, by itself, guarantee no transfers or establish compliance. The pilot should exclude unnecessary data and test the agreed restrictions.<\/p>\n<h3>Who owns the data, and what can we take away?<\/h3>\n<p>Your Company Brain belongs to you and remains installed on your infrastructure: you do not rent the Brain from Say Digital. At handover, you receive access, operating documentation and the delivered assets within the agreed scope needed for your team or another provider to take over. Export formats, dependencies and exit arrangements are documented in the contract. Third-party models and software remain subject to their licences; their ownership is not transferred to you.<\/p>\n<h2>Sources and limitations<\/h2>\n<p>Sources checked on 26 October 2026: <a href=\"https:\/\/aws.amazon.com\/what-is\/retrieval-augmented-generation\/\">AWS, explanation of RAG<\/a>; <a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/search\/retrieval-augmented-generation-overview\">Microsoft Learn, RAG and Azure AI Search<\/a>; <a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/search\/search-security-trimming-for-azure-search\">Microsoft Learn, security-filter pattern<\/a>; <a href=\"https:\/\/genai.owasp.org\/llmrisk\/llm062025-excessive-agency\/\">OWASP, LLM06:2025 Excessive Agency<\/a>; <a href=\"https:\/\/www.cnil.fr\/fr\/ia-et-rgpd-la-cnil-publie-ses-nouvelles-recommandations-pour-accompagner-une-innovation-responsable\">CNIL, AI and GDPR recommendations<\/a>. These sources document mechanisms and precautions, not a \u201cCompany Brain\u201d certification. The architecture, scenario and suggested tests are an editorial synthesis; they do not establish quantified gains or customer results. Compliance choices must be assessed for the processing activity concerned.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>A Company Brain turns scattered sources into verifiable business memory. A guide to use cases, permissions and testing, with Say Digital\u2019s proposed diagnostic, pilot and deployment service.<\/p>\n","protected":false},"author":2,"featured_media":14821,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_monsterinsights_skip_tracking":false,"footnotes":""},"categories":[94,164,172,84],"tags":[],"class_list":["post-13633","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-future-of-work","category-ia","category-ia-en","category-workflow"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v28.8 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>Company Brain: AI Knowledge Base for Your Business<\/title>\n<meta name=\"description\" content=\"Build a Company Brain from trusted business knowledge. Explore sources, access controls and AI, plus Say Digital\u2019s proposed diagnostic, pilot and deployment service.\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/say-digital.io\/blog\/company-brain-enterprise-memory-ai-agents\/\" \/>\n<meta property=\"og:locale\" content=\"fr_FR\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Company Brain: AI Knowledge Base for Your Business\" \/>\n<meta property=\"og:description\" content=\"Build a Company Brain from trusted business knowledge. Explore sources, access controls and AI, plus Say Digital\u2019s proposed diagnostic, pilot and deployment service.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/say-digital.io\/blog\/company-brain-enterprise-memory-ai-agents\/?lang=en\" \/>\n<meta property=\"og:site_name\" content=\"Say Digital I\/O\" \/>\n<meta property=\"article:published_time\" content=\"2026-09-27T12:38:13+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2026-10-26T23:25:27+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/say-digital.io\/blog\/wp-content\/uploads\/2026\/10\/company-brain-memoire-entreprise-agents-ia-cover.jpg\" \/>\n\t<meta property=\"og:image:width\" content=\"842\" \/>\n\t<meta property=\"og:image:height\" content=\"595\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/jpeg\" \/>\n<meta name=\"author\" content=\"Lia Amplify\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u00c9crit par\" \/>\n\t<meta name=\"twitter:data1\" content=\"Lia Amplify\" \/>\n\t<meta name=\"twitter:label2\" content=\"Dur\u00e9e de lecture estim\u00e9e\" \/>\n\t<meta name=\"twitter:data2\" content=\"18 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/say-digital.io\\\/blog\\\/company-brain-enterprise-memory-ai-agents\\\/?lang=en#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/say-digital.io\\\/blog\\\/company-brain-enterprise-memory-ai-agents\\\/?lang=en\"},\"author\":{\"name\":\"Lia Amplify\",\"@id\":\"https:\\\/\\\/say-digital.io\\\/blog\\\/#\\\/schema\\\/person\\\/8ee0600139c147afbd1fa3ced079c557\"},\"headline\":\"Company Brain: business memory for your team and AI\",\"datePublished\":\"2026-09-27T12:38:13+00:00\",\"dateModified\":\"2026-10-26T23:25:27+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/say-digital.io\\\/blog\\\/company-brain-enterprise-memory-ai-agents\\\/?lang=en\"},\"wordCount\":3531,\"image\":{\"@id\":\"https:\\\/\\\/say-digital.io\\\/blog\\\/company-brain-enterprise-memory-ai-agents\\\/?lang=en#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/say-digital.io\\\/blog\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/company-brain-memoire-entreprise-agents-ia-cover.jpg\",\"articleSection\":[\"Future of Work\",\"IA\",\"IA\",\"Workflow\"],\"inLanguage\":\"fr-FR\"},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/say-digital.io\\\/blog\\\/company-brain-enterprise-memory-ai-agents\\\/?lang=en\",\"url\":\"https:\\\/\\\/say-digital.io\\\/blog\\\/company-brain-enterprise-memory-ai-agents\\\/?lang=en\",\"name\":\"Company Brain: AI Knowledge Base for Your Business\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/say-digital.io\\\/blog\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/say-digital.io\\\/blog\\\/company-brain-enterprise-memory-ai-agents\\\/?lang=en#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/say-digital.io\\\/blog\\\/company-brain-enterprise-memory-ai-agents\\\/?lang=en#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/say-digital.io\\\/blog\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/company-brain-memoire-entreprise-agents-ia-cover.jpg\",\"datePublished\":\"2026-09-27T12:38:13+00:00\",\"dateModified\":\"2026-10-26T23:25:27+00:00\",\"author\":{\"@id\":\"https:\\\/\\\/say-digital.io\\\/blog\\\/#\\\/schema\\\/person\\\/8ee0600139c147afbd1fa3ced079c557\"},\"description\":\"Build a Company Brain from trusted business knowledge. Explore sources, access controls and AI, plus Say Digital\u2019s proposed diagnostic, pilot and deployment service.\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/say-digital.io\\\/blog\\\/company-brain-enterprise-memory-ai-agents\\\/?lang=en#breadcrumb\"},\"inLanguage\":\"fr-FR\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/say-digital.io\\\/blog\\\/company-brain-enterprise-memory-ai-agents\\\/?lang=en\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"fr-FR\",\"@id\":\"https:\\\/\\\/say-digital.io\\\/blog\\\/company-brain-enterprise-memory-ai-agents\\\/?lang=en#primaryimage\",\"url\":\"https:\\\/\\\/say-digital.io\\\/blog\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/company-brain-memoire-entreprise-agents-ia-cover.jpg\",\"contentUrl\":\"https:\\\/\\\/say-digital.io\\\/blog\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/company-brain-memoire-entreprise-agents-ia-cover.jpg\",\"width\":842,\"height\":595,\"caption\":\"A brain connected to information blocks on a pastel orange background\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/say-digital.io\\\/blog\\\/company-brain-enterprise-memory-ai-agents\\\/?lang=en#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Accueil\",\"item\":\"https:\\\/\\\/say-digital.io\\\/blog\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Company Brain: business memory for your team and AI\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/say-digital.io\\\/blog\\\/#website\",\"url\":\"https:\\\/\\\/say-digital.io\\\/blog\\\/\",\"name\":\"Say Digital I\\\/O\",\"description\":\"Un site utilisant WordPress\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/say-digital.io\\\/blog\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"fr-FR\"},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/say-digital.io\\\/blog\\\/#\\\/schema\\\/person\\\/8ee0600139c147afbd1fa3ced079c557\",\"name\":\"Lia Amplify\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"fr-FR\",\"@id\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/8817b2616d710fbc77dbbba6c455eebcdd7c8e299ff0dc15b9164e00ab16be69?s=96&d=mm&r=g\",\"url\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/8817b2616d710fbc77dbbba6c455eebcdd7c8e299ff0dc15b9164e00ab16be69?s=96&d=mm&r=g\",\"contentUrl\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/8817b2616d710fbc77dbbba6c455eebcdd7c8e299ff0dc15b9164e00ab16be69?s=96&d=mm&r=g\",\"caption\":\"Lia Amplify\"}}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Company Brain: AI Knowledge Base for Your Business","description":"Build a Company Brain from trusted business knowledge. Explore sources, access controls and AI, plus Say Digital\u2019s proposed diagnostic, pilot and deployment service.","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/say-digital.io\/blog\/company-brain-enterprise-memory-ai-agents\/","og_locale":"fr_FR","og_type":"article","og_title":"Company Brain: AI Knowledge Base for Your Business","og_description":"Build a Company Brain from trusted business knowledge. Explore sources, access controls and AI, plus Say Digital\u2019s proposed diagnostic, pilot and deployment service.","og_url":"https:\/\/say-digital.io\/blog\/company-brain-enterprise-memory-ai-agents\/?lang=en","og_site_name":"Say Digital I\/O","article_published_time":"2026-09-27T12:38:13+00:00","article_modified_time":"2026-10-26T23:25:27+00:00","og_image":[{"width":842,"height":595,"url":"https:\/\/say-digital.io\/blog\/wp-content\/uploads\/2026\/10\/company-brain-memoire-entreprise-agents-ia-cover.jpg","type":"image\/jpeg"}],"author":"Lia Amplify","twitter_card":"summary_large_image","twitter_misc":{"\u00c9crit par":"Lia Amplify","Dur\u00e9e de lecture estim\u00e9e":"18 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/say-digital.io\/blog\/company-brain-enterprise-memory-ai-agents\/?lang=en#article","isPartOf":{"@id":"https:\/\/say-digital.io\/blog\/company-brain-enterprise-memory-ai-agents\/?lang=en"},"author":{"name":"Lia Amplify","@id":"https:\/\/say-digital.io\/blog\/#\/schema\/person\/8ee0600139c147afbd1fa3ced079c557"},"headline":"Company Brain: business memory for your team and AI","datePublished":"2026-09-27T12:38:13+00:00","dateModified":"2026-10-26T23:25:27+00:00","mainEntityOfPage":{"@id":"https:\/\/say-digital.io\/blog\/company-brain-enterprise-memory-ai-agents\/?lang=en"},"wordCount":3531,"image":{"@id":"https:\/\/say-digital.io\/blog\/company-brain-enterprise-memory-ai-agents\/?lang=en#primaryimage"},"thumbnailUrl":"https:\/\/say-digital.io\/blog\/wp-content\/uploads\/2026\/10\/company-brain-memoire-entreprise-agents-ia-cover.jpg","articleSection":["Future of Work","IA","IA","Workflow"],"inLanguage":"fr-FR"},{"@type":"WebPage","@id":"https:\/\/say-digital.io\/blog\/company-brain-enterprise-memory-ai-agents\/?lang=en","url":"https:\/\/say-digital.io\/blog\/company-brain-enterprise-memory-ai-agents\/?lang=en","name":"Company Brain: AI Knowledge Base for Your Business","isPartOf":{"@id":"https:\/\/say-digital.io\/blog\/#website"},"primaryImageOfPage":{"@id":"https:\/\/say-digital.io\/blog\/company-brain-enterprise-memory-ai-agents\/?lang=en#primaryimage"},"image":{"@id":"https:\/\/say-digital.io\/blog\/company-brain-enterprise-memory-ai-agents\/?lang=en#primaryimage"},"thumbnailUrl":"https:\/\/say-digital.io\/blog\/wp-content\/uploads\/2026\/10\/company-brain-memoire-entreprise-agents-ia-cover.jpg","datePublished":"2026-09-27T12:38:13+00:00","dateModified":"2026-10-26T23:25:27+00:00","author":{"@id":"https:\/\/say-digital.io\/blog\/#\/schema\/person\/8ee0600139c147afbd1fa3ced079c557"},"description":"Build a Company Brain from trusted business knowledge. Explore sources, access controls and AI, plus Say Digital\u2019s proposed diagnostic, pilot and deployment service.","breadcrumb":{"@id":"https:\/\/say-digital.io\/blog\/company-brain-enterprise-memory-ai-agents\/?lang=en#breadcrumb"},"inLanguage":"fr-FR","potentialAction":[{"@type":"ReadAction","target":["https:\/\/say-digital.io\/blog\/company-brain-enterprise-memory-ai-agents\/?lang=en"]}]},{"@type":"ImageObject","inLanguage":"fr-FR","@id":"https:\/\/say-digital.io\/blog\/company-brain-enterprise-memory-ai-agents\/?lang=en#primaryimage","url":"https:\/\/say-digital.io\/blog\/wp-content\/uploads\/2026\/10\/company-brain-memoire-entreprise-agents-ia-cover.jpg","contentUrl":"https:\/\/say-digital.io\/blog\/wp-content\/uploads\/2026\/10\/company-brain-memoire-entreprise-agents-ia-cover.jpg","width":842,"height":595,"caption":"A brain connected to information blocks on a pastel orange background"},{"@type":"BreadcrumbList","@id":"https:\/\/say-digital.io\/blog\/company-brain-enterprise-memory-ai-agents\/?lang=en#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Accueil","item":"https:\/\/say-digital.io\/blog\/"},{"@type":"ListItem","position":2,"name":"Company Brain: business memory for your team and AI"}]},{"@type":"WebSite","@id":"https:\/\/say-digital.io\/blog\/#website","url":"https:\/\/say-digital.io\/blog\/","name":"Say Digital I\/O","description":"Un site utilisant WordPress","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/say-digital.io\/blog\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"fr-FR"},{"@type":"Person","@id":"https:\/\/say-digital.io\/blog\/#\/schema\/person\/8ee0600139c147afbd1fa3ced079c557","name":"Lia Amplify","image":{"@type":"ImageObject","inLanguage":"fr-FR","@id":"https:\/\/secure.gravatar.com\/avatar\/8817b2616d710fbc77dbbba6c455eebcdd7c8e299ff0dc15b9164e00ab16be69?s=96&d=mm&r=g","url":"https:\/\/secure.gravatar.com\/avatar\/8817b2616d710fbc77dbbba6c455eebcdd7c8e299ff0dc15b9164e00ab16be69?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/8817b2616d710fbc77dbbba6c455eebcdd7c8e299ff0dc15b9164e00ab16be69?s=96&d=mm&r=g","caption":"Lia Amplify"}}]}},"_links":{"self":[{"href":"https:\/\/say-digital.io\/blog\/index.php?rest_route=\/wp\/v2\/posts\/13633","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/say-digital.io\/blog\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/say-digital.io\/blog\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/say-digital.io\/blog\/index.php?rest_route=\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/say-digital.io\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=13633"}],"version-history":[{"count":6,"href":"https:\/\/say-digital.io\/blog\/index.php?rest_route=\/wp\/v2\/posts\/13633\/revisions"}],"predecessor-version":[{"id":14843,"href":"https:\/\/say-digital.io\/blog\/index.php?rest_route=\/wp\/v2\/posts\/13633\/revisions\/14843"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/say-digital.io\/blog\/index.php?rest_route=\/wp\/v2\/media\/14821"}],"wp:attachment":[{"href":"https:\/\/say-digital.io\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=13633"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/say-digital.io\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=13633"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/say-digital.io\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=13633"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}