Patient Memory for Healthcare AI

Give your AI agents persistent, consent-aware memory of every patient. Assembled from clinical records, claims, and conversations — governed by the patient's consent in real time.

Watch the Demo ↓
HIPAA Compliant
42 CFR Part 2
SOC 2 Type II
FHIR R4 Native
The Problem

Healthcare AI agents are amnesiac

A patient who describes symptoms on Monday is a stranger on Tuesday. Four different agents, four cold starts. No shared context, no memory, no continuity.

Care Navigator — Monday
"I see you have Type 2 diabetes. Can you tell me about your current medications?"
Diabetes Coach — Wednesday
"Welcome! Do you have any chronic conditions I should know about?"
Med Adherence Bot — Friday
"Hi there! What medications are you currently taking?"
ED Triage Agent — Saturday
No history of opioid use disorder. Standard pain management protocol initiated.

Every session starts from zero

Healthcare AI agents don't remember patients across sessions. They can't share context with each other. And the most dangerous gaps — like substance use history that requires separate consent — aren't just forgotten. They're invisible to the systems that need them most.

87%
of AI agent interactions start with zero patient context
4.2×
more likely to miss drug interactions without memory
23min
average time re-gathering info already shared
0
healthcare memory layers exist with consent built in
The SDK

Four primitives. One SDK.

Everything your AI agents need to remember patients, respect their consent, and prove what happened — in four API calls.

⬡
Patient Memory Graph
A persistent, evolving knowledge graph per patient — assembled from FHIR R4 clinical data, claims, pharmacy records, and AI agent conversations. Tracks temporal changes and resolves identity across channels.
const memory = await fabric.getPatientMemory({
patient_id: "pt_MCR_62",
agent_id: "care_navigator",
scope: ["conditions", "medications", "conversations"],
temporal: { lookback: "6months" }
});
// → Returns consent-filtered subgraph
◉
Consent-Aware Access
Granular, real-time consent enforcement at the memory layer. Supports HIPAA general consent, 42 CFR Part 2 segmentation, and state-specific rules. Per-patient, per-category, per-agent, and revocable.
// Consent is structural, not a filter
memory.nodes.forEach(node => {
node.consent_scope // who can see this
node.consent_authority // under what law
node.consent_expiry // auto-revocation
});
// 42 CFR Part 2 nodes: invisible by default
⊙
Access Audit Trail
Every memory read, write, and access decision is logged with full provenance. Who accessed what, when, under what consent authority, and what was returned — or withheld.
// Every query is logged immutably
{
agent: "ed_triage",
nodes_returned: 47,
nodes_withheld: 12, // 42 CFR Part 2
consent_scopes: ["general_medical"],
withheld_reason: "SUD_CONSENT_DENIED"
}
◈
Decision Traces
Captures what your AI agent did with patient memory and why. When a regulator asks "why did the AI recommend X?" — FabricAI provides the answer.
await fabric.logDecision({
patient_id: "pt_MCR_62",
agent_id: "ed_triage",
memory_used: ["conditions", "allergies"],
action: "pain_protocol_modified",
reasoning: "OUD history surfaced via consent"
});
Live Demo

See the difference memory makes

Maggie is a 62-year-old with Type 2 diabetes, CKD, and a history of opioid use disorder. Watch what happens when she arrives at the ED — with and without FabricAI.

MC

Margaret "Maggie" Chen-Rodriguez

62 · T2DM · CKD Stage 3a · HTN · MDD — ED Visit: Severe Abdominal Pain
Toggle AI Agent Mode
Compare how the same query performs with and without patient memory.
❌
No patient memory — starting from zero
Traditional AI has no memory of this patient. It cannot see prior conversations, cross-agent context, or protected SUD history. It may recommend opioids to a patient with OUD history.
OUD History
AUD History
Suboxone Rx
Naltrexone Rx
Detox 1997
Detox 1999
EHR OnlyTraditional AI Agent — No Memory
Patient presents with severe abdominal pain, onset 3 hours ago. Reviewing available records...
Active conditions: T2DM, CKD Stage 3a (eGFR 52), HTN, MDD
Current meds: metformin, insulin glargine, lisinopril, amlodipine, sertraline
Allergies: codeine (GI distress), statin intolerance
Recommending standard pain management: ketorolac IV (caution with CKD, short course only), acetaminophen IV as adjunct.

Consider oxycodone 5mg if pain uncontrolled.
⚠ DANGEROUS RECOMMENDATION — This patient has OUD history (in remission since 2013). The traditional agent cannot see this and may trigger relapse with opioid prescribing.
No substance use history available. No cross-agent context. No prior AI conversations.
Use Cases

Built for every healthcare AI agent

Whether you're building triage bots, care navigators, or chronic disease coaches — FabricAI gives them the memory and consent infrastructure they need.

Digital Health
AI Triage & Care Navigation
50K interactions/month, every session starts cold. Patients repeat symptoms, medications, and history. Critical context is lost between handoffs.
Today
"Can you tell me your medications?" — asked for the 4th time this month
With FabricAI
Agent knows conditions, meds, last 3 conversations, and patient preferences. SUD data auto-excluded from triage scope.
Payers
Care Management Outreach
AI outreach for chronic disease members. No memory of prior calls, no claims context, no pharmacy gap awareness. Members disengage.
Today
Generic outreach call with no context. Member hangs up in 30 seconds.
With FabricAI
Agent sees last 5 interactions, pharmacy gaps, recent ED visits. Picks up where the last agent left off.
Health Systems
Multi-Agent Coordination
Ambient scribe, portal chatbot, scheduling assistant — three agents that don't know each other exist. Context silos everywhere.
Today
Patient tells scribe about knee pain. Portal chatbot has no idea. Scheduling can't prioritize.
With FabricAI
All agents read/write the same memory graph. Knee pain → reduced activity → mood impact → glucose spike — connected.

"Every healthcare organization deploying AI eventually gets asked the same question: what did the AI know, and why did it do what it did? Today, nobody can answer that. We built the infrastructure so that patient memory, consent, and accountability are one system — not three separate problems."

Compliance

Built for healthcare from day one

Not a general-purpose memory layer with HIPAA bolted on. Every design decision starts with the regulatory requirement.

🛡
HIPAA
Privacy Rule, Security Rule, Breach Notification. Minimum necessary, patient access rights, accounting of disclosures.
🔒
42 CFR Part 2
Substance use disorder data segregated at the graph level. Separate consent per disclosure. Cannot be re-disclosed.
📋
SOC 2 Type II
Security, availability, processing integrity, confidentiality, and privacy. Continuous monitoring and annual audits.
✦
HITRUST CSF
The gold standard for healthcare security certification. Comprehensive risk-based framework mapped to 40+ regulatory standards.
Get Started

Give your AI agents the memory patients deserve

FabricAI is available in private beta for digital health companies, health systems, and payers building AI-powered patient experiences.

Watch the DemoRequest Access →