FAQ - questions enterprise clients ask

Nothing leaves. Sources are read into your own knowledge base; pages are rendered from it; Remy adds no database. The only outbound call is web search, and only where your security and compliance functions have approved it. → Architecture and data flow, Security, guardrails and compliance

No. Every run - interactive or scheduled - executes as the invoking user under their knowledge-base permissions and delegated connector permissions. A skill can never widen what its caller is entitled to see. Team and organisation views are bounded by folder permissions in the same way. → Connectors, identity and mining

No. Every letter is a draft in the user's own mailbox; the user sends. Core systems are read-only; KYC and custody fields are mirrored or blank, never inferred. Write-back is a separately governed decision outside the package. → Drafts and sending, Security, guardrails and compliance

The description file carries an exclusion list applied by every skill. It ships with politics, health, inferred personality, gossip, relatives' finances and tax law; compliance extends it without a software change. A user can also remove any fact with "forget". → Configuring the memory

Every fact carries a followable source, a date, a basis and a confidence; briefs show the basis beside every figure. Facts from a system of record are copied, never inferred. Anyone can correct a fact on the page or in chat; the change is logged and protected from automation. → Client memory - how it works, Working with a client's memory

Source references on every fact, an append-only change log per client, one alert ledger per monitor, a mail log that records "sent" only after the mailbox confirms it, the knowledge base's own versioning and audit log, and a read-only audit skill that reports memory health per client and per deployment. → Design phase, branding, language and audit

No. Authorship classes make it structurally impossible: mirror pages copy from the system of record or stay blank; agents propose; humans confirm. No screening, KYC, custody or performance figure is ever derived by the agent. → Client memory - how it works

The client memories are the bank's - knowledge-base folders under the bank's permissions, retained and backed up. Client documents and any team or organisation memory are untouched; a successor's first run rebuilds their own dashboard from the same memories on day one. → Rolling out to a team and scope levels

Yes. Scope applies to memories, skills and configuration at user, team and organisation level. A private-banking desk can mine and organise a relationship differently from an institutional desk; the contracts, provenance and audit trail are identical. Precedence: user over team over organisation, except what the bank locks. → Rolling out to a team and scope levels

The relationship manager / financial advisor, explicitly. Remy prepares the gap check and the hand-off pack; the lifecycle only moves on their confirmation. → Prospects, onboarding and offboarding

The morning page, the brief for every meeting in the coming week and a letter for every open item are ready at 07:00. The user reviews, decides and sends. → Why Remy

The run limits its conclusions, states which sources were used and which were unavailable, and produces an explicit blocked section or question rather than guessing. Nothing is lost; the next run reads everything since the last mining-log entry. → Running, monitoring, upgrading and recovery

The first run mines every client once, five in parallel; it scales with book size and document volume. A book of a few dozen clients is usually ready within the hour; large books are best started outside working hours. Everything after is incremental. → Your first day, Before you begin

Your own deployment of the platform with knowledge base, agent runtime, MCP hub, side panel and scheduled tasks; a frontier model; Outlook mail and calendar as the minimum connector; IdP federation for team scope; a technical-user token; a PDF renderer; named owners for each source system and for compliance, privacy and records. → Before you begin

Meeting briefs need calendar access; drift needs a portfolio connector; screening, market radar and referrals need approved web search; team and organisation scope need IdP group sync; PDF briefs need a renderer; branded output needs a brand pack. Each stays in the package and is activated once the dependency is in place. → Before you begin

Push the new skill tree (or take the new platform image); configuration lives in the knowledge base and is not overwritten; run the check skill; users pick up the version on their next run. Every service is versioned independently; rollback is pushing the previous tree. → Running, monitoring, upgrading and recovery

In your deployment of the platform - public cloud, private cloud or on premises. Hosting bounds model availability and outbound access, and therefore which web-dependent skills can run. → Architecture and data flow

A system attaches as an MCP server with a row in the identity map - no change to the memory. A skill is a prose task definition that raises findings through the same actions register and appears on the dashboard like the shipped ones, at user, team or organisation scope. No fork. → Extending Remy

Yes. Each monitor reads a policy file in the knowledge base; the scheduler retunes cadences from chat. Compliance owns the files; the knowledge base versions them. → Configuring the memory and tuning the monitors, Schedules

No. Remy is a contracted package, delivered complete and configured in a joint design phase so that skills, memory pages, monitors, scope and routing match your processes. After that, administrators configure and users install with one button. → Before you begin