Now available for alpha customers — AgentBrief for call centres

Know your member before
the call begins.

No model runs when a briefing is produced. A rule engine derives the facts and a fixed template renders them, so the same claims always produce the same briefing. The model runs upstream, before deployment, where a person reviews what it wrote before it ever runs.

AgentBrief turns the paid medical and pharmacy claims you already hold into a six-field, plain-language member briefing on your agent's screen in under 10 seconds — precomputed in batch inside your own AWS environment, so no inference sits in the call path.

Request a Demo See How It Works ↓
None
Model calls when a
briefing is produced
22/22
Fact statements correct in the
Phase 7 template comparison
6
Fields in the briefing,
none of them prose
<10s
Briefing on the
agent's screen
Patient data never leaves your AWS environment
AgentBrief is deployed inside your VPC, and producing a briefing calls no outside service and no model. By design, not by policy.
Model calls at briefing time
None
Data received by ArtiSoft
None
External API calls at runtime
Zero
The status quo

Your agents are flying blind.

Fragmented claims data means agents spend the first minutes of every call reconstructing a history the plan already holds.

Minutes lost before every call

Agents move between several disconnected systems to assemble a member's conditions, medications and recent care history before the conversation can start.

Pharmacy signals nobody sees in time

Missed refills, polypharmacy and adherence gaps sit in pharmacy data that member services and care management rarely have in front of them while the member is on the line.

High-risk members surface late

Without a risk flag on the record, outreach is reactive: a high-acuity member is identified after a gap has opened rather than while it can still be closed.

Quality measures pay for it

PDC adherence, CDC composite and TRC follow-up windows can be missed for no better reason than that the person on the call did not have the information in front of them.

How it works

On screen before the call connects.

AgentBrief reads the claims extracts you already produce and returns a structured member briefing — no new integration code, and no model in the call path.

01
Deploy inside your AWS environment
AgentBrief ships as a hardened, automated CloudFormation stack provisioned directly in your own account, and the deployment is customisable to your environment.
02
Unify medical, pharmacy and eligibility
Claims are reconciled, deduplicated and enriched into a single longitudinal member record, with every derived fact keeping a pointer back to the claim it came from.
03
Render the briefing deterministically
A rule engine derives the facts and a fixed template renders them into six short fields. No model is called at this step — local or remote — so the same claims always produce the same briefing.
04
Deliver to the agent screen
Briefings are precomputed in batch against your known contact list and surface in your existing call-centre platform — Genesys, Amazon Connect, Salesforce Health Cloud — before the call begins.
AgentBrief — member briefing Illustrative
Patricia L. — Member #04821
DOB 1957 · Medicare Advantage · PCP: Dr. Osei · Last claim: 3 days ago
High Risk 2 Open Gaps 3 Active Rx
Active conditions
Type 2 Diabetes (E11.9)Dx 2018
Hypertension (I10)Dx 2015
CKD Stage 3 (N18.3)Dx 2021
Flags derived from claims
Metformin 18 days past expected refill date — last fill Sep 14; PDC gap open. Source: pharmacy claim 9F-44120.
No HbA1c result on file since Mar 2024 — HEDIS CDC measure window open. Source: medical claims, 24-month window.
Rendered from claims — no model call
Where the model runs

The model moved upstream.

Removing the model from the briefing step did not remove it from the product. It moved to where its work can be checked: designing the rules and the template, before deployment, on a desk rather than on a call.

Before deployment — a frontier model drafts
A frontier model helps author the rules that derive a member's facts and the template that renders them. That is design work, and it is done once rather than per call.
Before deployment — a person reviews
What the model produced is read and corrected while it is still configuration. A briefing written at runtime can only be reviewed after the member has already heard it.
At runtime — nothing left to review
Producing a briefing runs no model at all: no prompt, no sampling, no inference cost and no outside dependency. Where the wider JetStore platform does run a model, for other workloads, it runs on compute inside your own VPC.
22/22
The template was correct on every runAcross 22 curated member fact sets, the deterministic template carried no statement untrue of the facts. The model carried one in 10 of the 22.
17/18
The model's error was one-directionalIn 17 of 18 member-runs where the member had no maintenance drug, the model called a non-maintenance drug a maintenance one — under a prompt flag that exists to prevent exactly that.
2/66
What only a model could do, it did twiceJoining a medication list to a diagnosis — the one thing a template cannot do — happened in 2 of 66 member-runs, and neither inference was licensed by the fact set.
0
Fields on which a model writes proseThe delivered briefing is six short fields. None of them is free text — by construction, rather than by a rate we are hoping stays low.

What was measured, and when. Phase 7 of ArtiSoft's JetStore agentic AI analysis (§8, read 2026-09-13) compared the deterministic template against a single 3-billion-parameter model over 22 curated member fact sets, in one comparison. That is strong evidence for a design decision and it is not a service level: your cohort, your feeds and your measures would have to be measured on your own data.

Built for your team

The right briefing
for every stakeholder.

The same deterministic briefing serves member services, care management and the people who have to approve the deployment.

Shorter calls.
Better conversations.

Handle time grows when an agent has to ask a member for a history the plan already holds. AgentBrief puts that history on the screen before the call connects — identically for every agent, on every call, because the renderer is deterministic.

None
Model calls
per briefing
<10s
Briefing on
screen
6
Fields, the same
six every call
Context before the first wordConditions, active medications and recent care are on screen when the call connects. No system navigation, no hold time.
Outreach lists you can orderBriefings are produced in batch against your contact list, so a queue can be ordered by what the claims show rather than by who called first.
The same briefing for every agentA new hire and a ten-year veteran see the same member picture in the same words, because a template does not vary with the reader or the run.
Fewer repeat contactsAn agent who can see the member's claims history is less likely to end the call with a question still open.

Close more gaps.
In fewer calls.

Care managers are clinicians who spend a large part of the day assembling data. AgentBrief hands them the assembled version — conditions, adherence signals and open quality-measure gaps — before they dial.

The briefing is informational. It reports what the claims show, names the claim each line came from, and recommends no clinical action; the clinical judgement stays with your staff, where it belongs.

Adherence gaps while the year is openA PDC gap appears on the briefing with the fill dates it was derived from, in time for the conversation to matter.
Open quality-measure flagsCDC, CBP, SPC and TRC gaps are listed per member as claims-derived facts — stated, not interpreted, and with no suggested clinical action.
Recent inpatient activity, surfacedA recent inpatient claim appears as a flag, so transitions-of-care outreach can happen inside the follow-up window rather than after it.
Every line traceable to a claimEach field names the claim it was derived from — which matters when a representative restates the briefing to the member on a recorded line.

Your data never leaves
your environment.

AgentBrief is deployed entirely within your AWS infrastructure. Producing a briefing calls no external service and no model at all; where the wider platform does run a model, it runs on compute inside your VPC. No patient data is transmitted to ArtiSoft or to any third party, so your compliance perimeter stays exactly where it is today.

Fully automated deployment.

AgentBrief's stack is deployed with AWS CloudFormation, so the whole footprint is reviewable as code before anything is provisioned.

No PHI egress — by designClaims are processed entirely within your own AWS account. ArtiSoft receives no data, holds no data, and has no access to your environment at runtime.
Nothing for a BAA to coverBecause ArtiSoft never receives or processes PHI, there is no data-sharing relationship to paper over. Your counsel will want to confirm the conclusion for your own deployment; the clean boundary is the point.
Deployed as infrastructure-as-codeA hardened CloudFormation stack with least-privilege IAM scoping. Your team reviews and approves every permission before anything runs — no wildcard roles, no cross-account access.
Flat files or FHIR-shaped dataAgentBrief reads the claims extracts you already produce, against your canonical model. Your team writes no integration code.
Security & compliance

Your data stays
in your hands.

AgentBrief never sees your patient data. It runs inside your AWS environment, produces a briefing without calling any external service, and writes its logs to your own account.

HIPAA
No PHI reaches ArtiSoft
Provenance per field
Lineage preserved across claim files
Your data never leaves your AWS environment
AgentBrief is deployed inside your own VPC. Briefings are rendered on compute you control, and no patient data is transmitted to ArtiSoft or to any external system.
Inside the compliance scope you already maintain
AgentBrief runs in your environment, so it sits within the controls and the audit scope you already operate. We support customers who maintain HITRUST CSF certification. ArtiSoft itself holds no HITRUST certification and does not need one — certification covers an environment that handles the data, and no PHI is transmitted to us, so there is no ArtiSoft-side scope for it to cover.
No BAA to negotiate — ArtiSoft receives no PHI
Because ArtiSoft never receives or processes PHI, there is no data-sharing arrangement for a Business Associate Agreement to govern: you are deploying software rather than sharing data. Your legal team should confirm the conclusion for your own deployment.
Deployed as hardened infrastructure-as-code
AgentBrief ships as a reviewed CloudFormation stack with least-privilege IAM scoping. Your team approves every permission before deployment — no wildcards, no cross-account roles, no surprise egress.
ArtiSoft has no visibility into your environment
Audit logs, access events and generated briefings are written to your own CloudWatch or SIEM. ArtiSoft cannot monitor or read your data; support is delivered through your team, not ours.
Minimum-necessary projection, and provenance per field
A briefing carries the fields the call needs and no more, and each field names the claim it was derived from. These are the two guardrails that a deterministic renderer still needs — the second because a representative may restate the briefing to the member on a recorded line.
Three of the original five guardrails no longer have a subject
Five guardrails were designed for a briefing a model wrote. Once the renderer became deterministic, three of them had nothing left to check: a template that never writes prose cannot invent a fact, drift from the schema, or say something the fact set does not support. We list the two that remain rather than the five that were drawn.

See AgentBrief run against
your member population.

Book a 30-minute walkthrough. We will show you a briefing rendered from a synthetic cohort that mirrors your plan's demographics and condition mix — and the rules that produced every line of it.

Request a Demo Ask for the Technical Overview
Contact

Have a question? Let's talk.

Email us for more information about AgentBrief, deployment timelines, or to start a conversation with our team.

info@artisoft.io