Memory Fabric
SCRIMED memory is scoped, auditable, and separated by session, operational, and knowledge layers.
SCRIMED AgentOS v1 orchestrates synthetic pilot and enterprise assessment workflows only. It does not autonomously diagnose, treat, submit payer transactions, contact patients, file clinical documentation, or execute live care operations.
session
Session Memory
Short-lived task context for a single synthetic pilot request, workflow review, or assessment session.
Scoped to assigned agent, reviewer, and tenant role.
- Ephemeral by default; production retention requires tenant policy and audit approval.
- Prohibited: PHI, patient identifiers, live clinical records, payer member identifiers
- Audit: session memory opened, session memory summarized, session memory cleared
operational
Operational Memory
Workflow state, routing decisions, approval dispositions, exception reasons, and synthetic pilot outcomes.
Tenant-scoped operators, governance owners, and approved reviewers.
- Durable only for synthetic pilot evidence and approved enterprise assessment records.
- Prohibited: unapproved patient data, raw chart text, unredacted claims payloads
- Audit: workflow state recorded, approval disposition stored, exception reason linked
knowledge
Knowledge Memory
Guidelines, protocols, policies, publications, connector contracts, and governance source references.
Read-only for agents; write access requires knowledge steward and TrustQA verification.
- Versioned with validation timestamp, source owner, and retirement policy.
- Prohibited: uncited clinical claims, unversioned policy excerpts, unapproved local protocols
- Audit: source registered, source version validated, source retired
RBAC permissions
Least-privilege access is required before live healthcare use.
Tenant configuration, pilot scope, users, connectors, and governance controls.
Allowed: manage tenant settings, configure synthetic workflows, assign reviewers, view audit summaries
- Denied: bypass TrustQA, submit live clinical actions without production controls, delete immutable audit traces
Clinical review queues, documentation drafts, care-gap prompts, and evidence-linked recommendations.
Allowed: review clinical outputs, approve or reject draft recommendations, request evidence clarification
- Denied: configure connectors, change governance policy, approve payer submissions outside assigned workflow
Prior authorization, denial-risk, claims, appeal drafts, and payer policy evidence.
Allowed: review payer packets, approve RCM drafts, request missing evidence
- Denied: finalize clinical diagnosis, override clinical review, submit payer transactions without approval
Scoped agent execution inside sandboxed synthetic or approved tenant workflows.
Allowed: read assigned memory, write audit events, generate review-only outputs
- Denied: read unrelated tenant memory, perform autonomous live actions, modify RBAC policy