Vectyr builds custom AI systems that read where our clients' work already happens, turn it into verified structured truth through layered extraction, and drive the programme views, registers, reporting, and calculated results their business runs on. This is the documentation behind that: how we protect the data, which models are permitted to see it, and where our program has gaps.
This is our general documentation. Anything specific to an engagement — its architecture, connector scopes, model register, or evaluation results — is shared with that client directly under NDA and never published here, which is equally true of every other client.
We do not hold a SOC 2 report, and we lead with that rather than burying it. The reason is scope: a SOC 2 opinion covers a defined system that has been operating, and we build bespoke software per client, so at the point you are reviewing us that system is being designed rather than running. We build to the Trust Services Criteria from the first commit, submit the delivered system to examination once it is live, and decline to offer substitutes that resemble assurance without being it. See exactly where we stand.
Six commitments govern every system we build. The confidential documents are the evidence for them, including the places where a commitment depends on which deployment model you choose.
Client data is processed by models inside an agreed boundary: an enterprise endpoint under no-training, zero-retention terms, a model inside your own cloud tenancy, or open-weight models on compute you control. No client data trains any model.
Each client has its own database, its own object storage, and its own credentials. There is no shared application datastore, so there is no query that could return another client’s data.
Connections use an app registration you create, consent to, and can revoke without contacting us. We ask for the narrowest scope that works — site-by-site rather than whole-estate.
TLS 1.2 or better in transit, AES-256 at rest. Credentials are held as Cloudflare Workers secrets, injected at runtime, never in code or logs, and rotated on personnel change and on any suspected exposure.
Every assertion links to the meeting, message, or document it came from, with a timestamp and the person who confirmed it. Every change and override stays attributable.
Consequential updates are proposed and confirmed by a named person, not applied autonomously. Our systems observe and advise; they never control physical or safety-critical operations.
Readable by anyone, no NDA required. These are summaries; the detail is in the confidential library.
What Vectyr builds, how we protect it, and how to run a review with us.
How Vectyr protects client data across the systems we build and operate, and what a reviewer should expect from us.
Where Vectyr systems run, how each client's data is kept separate, and the two options for how AI models process it.
Which models process client data, what they are permitted to see, how outputs stay explainable, and why our systems are not black boxes.
The components of a Vectyr system, how data moves from source to verified record, and where the trust boundaries sit.
A practical route through a Vectyr assessment, what we will hand over, and what we need from you.
Classification, retention, subprocessors, and privacy commitments.
What personal data our systems touch, the legal basis and roles involved, retention, and the rights we support.
The third parties that may process client data on our behalf, what each does, and how we handle changes.
Incident response, recovery objectives, backups, and availability.
How to report a security issue to Vectyr, what is in scope, and the safe-harbour terms we offer researchers.
Control mappings and completed questionnaire responses for procurement.
Why a SOC 2 report cannot precede the system it would cover, what we build to instead, and the honest status of everything else.
38 documents covering our full policy set, architecture detail with exact integration permission scopes, the AI model register, resilience plans with recovery objectives, completed HECVAT and CAIQ responses, our control framework crosswalk, and sector annexes for research institutions and critical infrastructure. Everything downloads as PDF.
Access takes about a minute: tell us who you are, accept a short NDA, and click the link we email to your work address.
Request access Read the NDA first
The policies and standards that govern how we build and operate.
The governing policy for Vectyr's security program — scope, principles, roles, and the standards it delegates to.
How identities, authentication, authorization, and privileged access are managed for Vectyr staff, end users, and service credentials.
Cryptographic controls in transit and at rest, secret handling, rotation, and who holds keys in each deployment model.
How code is written, reviewed, tested, and released, and the security requirements that apply at each stage.
How changes to production systems are authorized, deployed, verified, and reversed.
How vulnerabilities are identified, triaged, and remediated, with target timelines by severity.
What we log, what we deliberately do not log, how long records are kept, and how clients get audit evidence.
How the devices and physical working arrangements of a fully remote team are secured.
Screening, confidentiality obligations, security training, and the onboarding and offboarding process for anyone with access.
The rules governing how Vectyr personnel may use client data, AI tools, and company systems.
Where systems run, how data moves, and which deployment model applies.
Component-level architecture, per-model egress paths, data stores, and the full set of network flows.
How client data is kept separate, how separation is enforced and verified, and what happens at the edges.
Exact Graph permission scopes, consent model, authentication method, and controls for Teams, SharePoint, and OneDrive connections.
Supported federation protocols, our SAML 2.0 and Shibboleth readiness profile, provisioning, and multi-factor inheritance.
Every integration type we build, the data it reads, the permissions it needs, and the questions to settle before connecting it.
Which models we use, what they may process, and how humans stay in control.
The models and endpoints approved to process client data, their retention and training terms, and the rules for changing them.
The data model behind every traceable assertion, how confirmation works, and what a reviewer can independently verify.
The failure modes of the AI systems we build, their consequences, and the controls that reduce them.
How extraction quality is measured, what we report, and how prompt and model changes are gated.
Classification, retention, subprocessors, and privacy commitments.
Our classification tiers, how they map to client schemes, and the handling rules and deployment requirements for each.
Default retention by data category, deletion on request, and what happens to data when an engagement ends.
Where data is processed, transfer mechanisms, and how residency requirements in Europe and Asia are satisfied.
The security and privacy terms Vectyr will commit to contractually, including notification timelines and audit rights.
How we assess, approve, monitor, and remove third parties that touch client data.
Incident response, recovery objectives, backups, and availability.
How Vectyr identifies, assesses, treats, and reviews information security risk.
How Vectyr detects, classifies, contains, and communicates security incidents, with committed notification timelines.
Recovery objectives, scenario-by-scenario recovery plans, and how continuity is assured for a small supplier.
What is backed up, how often, where it is held, how long it is kept, and how restores are tested.
What we commit to on uptime, support response, and maintenance, and what we deliberately do not commit to.
Who builds and operates your system, and the measures that keep it yours regardless of what happens to us.
Control mappings and completed questionnaire responses for procurement.
Vectyr's controls mapped to SOC 2, ISO/IEC 27001 Annex A, and NIST CSF 2.0, with an implementation status for each.
Vectyr's narrative response to the Higher Education Community Vendor Assessment Toolkit, by domain.
Vectyr's response to the Cloud Security Alliance Consensus Assessments Initiative Questionnaire, by control domain.
How we test, the roadmap for independent testing, and how to commission your own at no charge.
Our conformance against WCAG 2.1 Level AA, how we test, and how we handle accessibility review.
Corporate details and the diligence artifacts procurement teams normally request.
Additional controls for research institutions and critical infrastructure.
Additional controls for universities, academic medical centres, and research departments, covering human subjects, health data, student records, and sponsored research.
Additional controls for ports, terminals, and industrial operators, covering the operational technology boundary, safety, and sensitive operational data.
Security issues and vulnerability reports: trust@vectyr.co. We acknowledge within one business day. See our disclosure policy for scope and safe-harbour terms.
Questionnaires, agreements, and procurement: trust@vectyr.co. Send us your questionnaire in your own format and we will complete it.
