Coding agent
Proposes and justifies ICD-10/CPT with a denial-risk score (Codexa).
Agentic AI
Agents that finishthe encounter.
Scribara is agent-native. A planner, executor, verifier, and writer collaborate to complete coding, prior auth, referrals, and follow-up — taking real action, always under clinician oversight.
Reliability is engineered — verifier models, evals, and guardrails
Epic FHIRathenahealthNVIDIA InceptionHL7SOC 2HIPAA
Plan
Scope the work.
The planner reads the encounter and decomposes it into note, codes, auth, referral, and follow-up.
Reason
Ground every step.
The executor drafts each output grounded in guidelines and payer rules via NeMo Retriever.
Verify
Catch the failures.
An independent verifier checks outputs; NeMo Guardrails enforce abstention and escalation.
Act
Only with sign-off.
The writer executes approved actions and records an immutable, reversible trail.
Agents
The Scribara agent fleet
Specialized agents that take real financial and administrative action.
Prior-auth agent
Assembles, submits, and tracks payer authorizations (Authra).
Referral agent
Writes specialty referral letters from the encounter.
Follow-up agent
Schedules and drafts follow-up; bridges toward chronic-care. [ASPIRATIONAL]
Verifier agent
Independent check trained on real failure modes.
Appeals agent
Drafts evidence-based denial appeals. [ASPIRATIONAL]
Guardrails
Autonomy you can defend
Consequential actions require clinician approval. Budget caps, escalation paths, and two-person rules are first-class — exactly the controls regulated buyers demand.
- Abstain or escalate when confidence is low
- Human-in-the-loop on every consequential action
- Immutable audit of every step and tool call
- Per-action permissions and approval policies
Planscoped
Verifypassed
Escalate1 item
Actsigned
Reliability
Trust, quantified
0
Consequential actions reviewed
0
Independent verifier
0
Silent failures by design
0
Specialist agents
Observability
Every decision is inspectable
Drill from any KPI down to the exact agent action, tool call, and evidence — readable as plain text, not trapped in motion.
# Agent run — cardiology encounter
plan note, codes, prior_auth, referral
perceive asr=ok imaging=n/a ehr=loaded
reason codes=["I25.10","93306"]
verify "93306" confidence=0.71 → escalate
act awaiting clinician sign-off ✓
Proof
What clinicians say
Outcomes are reported by design partners and marked illustrative [PLACEHOLDER].
“It does the work and shows the receipts — that's why we trust it.”
“Escalation when unsure is the feature that won over our compliance team.”
Trust
Enterprise-grade from day one
Healthcare buys on trust. Scribara is built to clear procurement, security review, and compliance.
HIPAA / BAA
PHI handled under a Business Associate Agreement.
SOC 2 Type II
[ASPIRATIONAL] — roadmap to Type II within 12 months.
On-prem NIM
Run owned models inside your datacenter via NVIDIA AI Enterprise. [ASPIRATIONAL]
Immutable audit
Every agent action logged, reversible, tied to the clinician's signature.
Data residency
US, EU, UK regions; per-tenant keys (BYOK option). [ASPIRATIONAL]
ISO 42001
AI management-system conformity on the roadmap. [ASPIRATIONAL]
Answers
Frequently asked
No consequential action ships without clinician approval; low-confidence steps abstain or escalate.
An independent verifier and NeMo Guardrails enforce abstention, and outputs are grounded via retrieval over guidelines and payer rules.
Yes — an immutable, reversible audit logs every step, tool call, and the evidence behind it.
Put agents to work on the busywork.
Watch the fleet complete a real encounter, end to end, with sign-off.