Answers for the closed beta.

Substrate is a memory layer for agents. It turns supported, authorized context into canonical memory about people, organizations, products, projects, commitments, relationships, and change. The current foundation can publish searchable, cited memory within a verified deployment boundary; the closed beta is testing the customer workflows and outcomes built on that foundation.

No. Those systems may hold source context or remain the system where work is performed. Substrate is intended to preserve qualified memory across agent workflows. It does not replace operational systems, human judgment, or a team’s authority model.

No. It is deliberately bounded by available sources, authorization, publication rules, retrieval scope, and relevance to a durable entity. It can miss context, interpret it incorrectly, or remain uncertain. The product is designed to keep evidence and qualifications inspectable rather than claim complete or infallible memory.

No. Memory can support preparation, drafting, or a decision. It does not grant permission to send a message, update another system, commit resources, or act on someone’s behalf. Application authority remains with the integrating workflow and its human or system controls.

The GTM beta is designed to test whether cited organizational memory improves continuity and customer judgment in a real customer-facing agent workflow. A candidate first-value event is one useful, evidence-backed change that a participant would otherwise have missed or reconstructed, followed by inspection of the memory and provenance. This is an evaluation goal, not a published outcome claim.

It is a separate, concierge-led beta in development. The proposed workflow would test a selective briefing around people, commitments, and material change. It is not a currently available self-serve product mode, a complete account of a founder’s company or life, or an autonomous assistant.

No. The initial topology is one invited participant per isolated environment, acting for one defined organizational unit. Collaborative access, shared multi-tenancy, team permissions, comments, and simultaneous editing are not current beta claims.

A strong partner already has a consequential agent workflow, can authorize a bounded source set, and will review identity, citations, freshness, conflicts, and corrections—not only whether an answer sounds plausible. The initial beta is managed and selective rather than open registration.

The implemented foundation supports specific retained conversation context, explicit memory writes and captures, supported documents, operational traces, and existing canonical memory. Calendar, CRM, support, email, and product-usage sources are potential managed-beta inputs, but availability varies by deployment.

Onboarding names the actual supported sources and authority boundary. A broad connector catalog is not promised.

An observed or explicit claim is supported directly by an authorized source. An inference is an evidence-bounded interpretation and should remain labeled as such. Assistant suggestions, generated summaries, and tool output do not independently establish facts. When support conflicts or remains weak, the memory can stay conflicting or uncertain instead of being promoted to certainty.

A citation provides a route to the authorized source or provenance that supported a memory. It does not prove the source was truthful, complete, current, correctly interpreted, or authorized for every future task. Citation is an inspection aid, not a correctness seal.

Time and status stay part of the memory model. Current context is intended for present use; historical context explains prior states; superseded context has been displaced by newer support. Conflicting, uncertain, and inferred context can remain visible so retrieval does not silently flatten them into one story.

Source replacement, provenance, identity decisions, and revision state provide a basis for correction. A challenge should preserve the evidence and history rather than simply overwrite the visible sentence. The underlying mechanisms exist, but the complete self-serve customer correction journey remains closed-beta work.

Each initial participant receives an isolated environment whose corpus, identity, credentials, jobs, logs, backups, support path, export, and deletion path belong to that deployment. Current VM deployments use an external authenticated gate; the application is not represented as a complete shared customer identity or role-based access system. See the security page for the implemented controls and limitations.

No blanket assurance is made. Isolation, authenticated ingress, private evidence storage, publication checks, credential containment, and deletion workflows are specific controls with operational assumptions. No certification or universal compliance claim is represented. The exact provider, support-access, retention, and security boundary must be confirmed for each deployment.

Retention is deployment- and data-specific. The operational schedule must name source retention, raw event retention, logs, jobs, support records, provider handling, backups, and legal or contractual requirements. The beta should not be used for regulated or unusually sensitive material unless it has been separately reviewed and explicitly accepted.

Active deletion is modeled at the source contribution and evidence boundary. Affected claims and relationships are suppressed or removed, published pages and search surfaces are deleted or rebuilt, and remaining memory is recomputed from safe support. Active-service deletion does not itself claim immediate erasure from every backup, external provider, or human-controlled copy; those procedures must be specified for the deployment.

No. Do not submit passwords, tokens, private keys, production credentials, or sensitive datasets through the public application or ordinary email. The beta team should agree an approved transfer method and source boundary before any evaluation data is provided.

No. The present service exposes managed integration surfaces inside the closed beta, but public SDK names, endpoints, authentication, quotas, scopes, and compatibility commitments are not final. Examples on the developer page are illustrative contracts, not released specifications.

An integrating workflow sends supported, authorized context and retrieves bounded published memory for a named subject, agent, scope, and task. The workflow remains responsible for its own permissions, tools, decisions, and external actions. Substrate is intended to be a memory layer beneath existing agent work rather than a replacement for the application.

Bring one consequential workflow, the source categories you can lawfully and appropriately authorize, the person responsible for reviewing results, and the criteria that would make the memory useful or unsafe. Expect a managed deployment, evolving interfaces, explicit failure states, and hands-on quality review.

Submit the closed-beta application with a real memory problem, your current agent workflow, the source categories you could authorize, and how you would evaluate the result. Select either the GTM path or the managed founder design-partner path; selection is not guaranteed.