Governed AI & company decision records
Company decisions can contain sensitive plans, personal details, and competitive knowledge. Reality Skill combines a protected provider route with local redaction controls and a permission-scoped record of what AI received, what it returned, and how people used it.
Reduce unnecessary exposure. Preserve a reviewable trail. Keep the decision human-owned.
Two layers of protection
Sending less sensitive information and keeping a trustworthy record are different jobs. Reality Skill is designed to do both for company-owned decisions when the organization policy is enabled.
AWS states that Bedrock inputs and outputs are not shared with model providers and are not used to train base models. Reality Skill’s current Bedrock route has model invocation logging off and the configured account retention setting at none.
Read AWS’s Bedrock privacy statementCompany policy can require a local scan for configured identifiers, credentials, and protected terms, followed by redaction, a second scan, an exact outbound preview, and human approval before dispatch.
Redaction reduces exposure. It cannot guarantee that every secret or identifying detail has been removed from free text.
Permission-scoped records can preserve the approved outbound payload, AI response, model, prompt and policy versions, approval, payload hash, and transport status. Security and quality reviewers can trace the influence without relying on memory.
These company audit records are intentionally retained. They are not provider training logs and are not described as “all logs cleared.”
The practical comparison
The avoidable risk is often unmanaged copy-and-paste: staff use personal AI accounts, settings vary, sensitive context leaves the normal process, and the company cannot later reconstruct what shaped the recommendation.
Reality Skill adds the decision context around the model: an approved route, a reviewable outbound payload, accountable human judgment, and a record connecting AI input to the final decision and outcome.
Ownership and continuity
Free personal cases remain owned by the individual account. Company-owned cases are separate: an accepted company decision lead provides continuity, former members lose access when they leave, and the record remains available to permitted current participants. Company custody applies only to cases explicitly created or transferred under the enabled company workflow.
Team visibility
Structured Team contributions follow the stated room visibility rules; independent input alone is not anonymity. Separately, current invited participants can add protected feedback with named or anonymous-to-team visibility. Anonymous notes omit the author from company-facing views and supported exports, including views for leaders and company admins. Only the author can voluntarily reveal a note. During initial collection notes remain sealed; after successful reveal, new notes become available on refresh.
Protected feedback
Infrastructure operators retain technical database access. Content, small-team context, and watching changes can suggest identity. Hosting-log and incident-operation reviews remain outstanding. Eligible company moderators can hide abuse, private information, or spam with recorded reasons; a placeholder and history remain, and participants can request reconsideration. Moderator access to preserved text does not include the anonymous author mapping. Previously downloaded copies cannot be recalled. No guarantee against retaliation or claim of whistleblower-system compliance is made.
Optional AI routes
The manual decision workflow remains available without AI. AWS Bedrock is the current default route. AWS states that Bedrock inputs and outputs are not shared with model providers and are not used to train base models; Reality Skill’s current route has the configured account retention setting at none and model invocation logging off. A user can explicitly select the current OpenRouter alternative for personal work. There is no automatic AWS-to-OpenRouter fallback. Managed Bedrock is not a dedicated customer VPC deployment.
Company AI evidence
For company-owned decisions, a versioned company policy can require a local identifier, credential, and protected-term scan; redaction and a second scan; an exact outbound preview; human approval; and append-only records of the approved payload, response, model, prompt and policy versions, payload hash, and transport status. These permission-scoped records support security and quality review. They are intentionally retained rather than described as cleared. Redaction reduces exposure but cannot guarantee that every secret or identifying detail is removed.
Usage and records
Decision content is not used for advertising, model training, or cross-user benchmarks. Personal Vault exports and company review packets cover supported records within the caller’s permissions; neither promises every operational data category. Retention differs between cases, packets, receipts, and restricted backups.
Procurement questions
Ask us to confirm required authentication, integrations, hosting, access, export coverage, custody, and support arrangements in writing. This page does not represent a certification or a commitment to support every enterprise requirement. Contact us with the requirements first, without sending confidential decision records.
Questions, answered
No. Personal cases remain individually owned. Company custody applies only to decisions explicitly placed in the enabled company workflow, with its own decision lead, permissions, and continuity rules.
Yes. AI is optional, and the manual decision workflow remains available. Invoking AI sends relevant material through the providers described in our privacy notice.
Not through the company interface or supported exports. Only the author can voluntarily reveal a protected note. Moderators can review hidden original text without receiving its private authorship mapping. This is anonymity from the team, not from infrastructure operators; content and context can still suggest identity.
Team Room protected feedback is scoped to authorized invited participants. Company Consultation is a separate workflow for 8–100 invited company members, with named or anonymous-to-audience contributions. It broadens input without expanding the core decision team or granting decision authority.
No. Provider-side model invocation logging is off on the current Bedrock route, but Reality Skill intentionally preserves permission-scoped company evidence when the enabled policy requires it. That evidence can include the approved outbound payload, response, model and prompt identity, approval, payload hash, and transport status so authorized security and quality reviewers can trace what influenced a decision.
Reality Skill gives the company a consistent route, policy, redaction and approval controls, and an auditable connection between AI assistance and the final decision. The comparison is with unmanaged personal-account use, where settings and records can vary.
No. Bedrock is the current managed default route with the configured retention and logging controls. A dedicated or company-specific private route is a separate requirement that must be assessed and agreed.
The Privacy and AI Data Use page explains the operator, data use, sharing, retention, exports, requests, and website analytics. Use the website contact form for organization-specific questions.
For your leadership team
See the platform in action, discuss an organization license, and identify an initial decision your team can work through.
Request a platform demo