Carelytic has been an AI-first platform since day one. But until now, the AI showed up as a handful of separate features — a SOAP-draft button here, a coding-review panel there — each doing its own thing. We've unified all of it into a single engine with a single name: Sage.
Sage is one AI brain. Every AI surface in Carelytic is a facet of it, which is why they all behave the same way: grounded in your real data, gated by the same permissions as your staff, and getting smarter from the same shared learning loop. This post walks through what Sage does today, and — just as important for a HIPAA platform — how it's built so you can trust it.
What Sage does
Sage shows up wherever the work happens, not in a separate "AI tools" corner:
- Ask Sage — a conversational drawer on every page. Ask a how-to, look something up, or hand Sage a task. It answers using tools scoped to exactly what you're allowed to access.
- AI Reports — describe the report you want in plain English ("LUPA-risk visits by branch this month") and Sage builds a live, tenant-scoped report plus a written summary of what it found.
- Audit AI Search — "Who logged in yesterday?" "Who viewed this patient's chart last week?" Sage turns the question into a permission-gated audit query and answers with the real names and rows.
- AI Document Intake — drop a referral packet or a face sheet on the new-client or employee wizard. Sage reads the PDFs and images and pre-fills the form field by field. Nothing saves until an admin reviews and confirms each value.
- Clinical drafting — SOAP draft from the structured findings a clinician already entered, Pre-Sign QA against CMS Conditions of Participation, Carry-Forward from the prior visit (findings only, never vitals or signatures), and the CMS-485 narrative sections drafted from the locked OASIS.
- Sage for billing — a pre-bill claim scrub that explains what would reject before you transmit, and a denial analyst that reads a denied claim and recommends appeal, correct-and-resubmit, or write-off. Advisory only — Sage never edits or submits a claim itself.
The learning loop: correct Sage once, and it remembers
Here's the part we're proudest of. When a biller, DON, or coder catches Sage getting something wrong for your agency, they don't file a ticket and wait for a model update. They teach Sage — a one-click correction from wherever they found the problem. It becomes a proposed Lesson.
A Lesson does nothing until a Carelytic admin approves it. Once approved, Sage applies it instantly the next time the situation comes up — no retraining, no fine-tuning, no code change. Because it's retrieval rather than retraining, every Lesson is reversible: pause or archive it and it stops applying immediately. A thumbs-up or thumbs-down on any Sage answer feeds the same loop.
And when a "correction" turns out to be a genuine hard rule — "payer X always needs modifier Y" — it gets promoted out of the AI prompt and hardened into enforced configuration, where it applies 100% of the time. AI is for judgment; rules belong in code.
Built to stay on the right side of the line
The HHS Office of Inspector General and the DOJ-HHS working group have been explicit that AI tools recommending coding changes purely to maximize reimbursement are a fraud-enforcement priority — the line the recent home-health False Claims Act settlements were drawn on. Sage is built to live on the right side of it, and to keep your agency there too:
- Evidence first. Every coding finding cites the chart text it came from. Sage surfaces documentation-specificity gaps and doc-versus-code mismatches — including where your documentation supports a more precise code and where it doesn't. It does not recommend changes purely to lift reimbursement.
- Human sign-off. Sage never auto-applies a coding change, never edits a claim, never submits. A person reviews every recommendation and signs.
- Grounded, not guessing. For anything touching patient data, Sage picks from a validated whitelist and the app runs the real, tenant-scoped query. Your data comes from your database, never from the model's imagination.
- Same gates as your staff. Every Sage query runs through the exact permission-gated code path your team uses. Sage can't reach data the person asking couldn't already reach.
- An outage is never silent. If Sage is ever unavailable, the workflow falls back to the manual path and your admins are alerted.
It maintains itself as your data grows
The hard part of clinical AI isn't the first demo — it's staying correct as an agency's data grows. What Sage can see is derived from the data models at runtime, so when you add a new field to a chart or a report, it flows to Sage automatically. Nobody has to remember to "tell the AI about it." That's the single biggest reason Sage doesn't quietly drift out of date.
Included in every tier
Sage runs on a frontier LLM (Anthropic's Claude), right-sized per task to stay fast and affordable, under the same HIPAA-aware architecture, RBAC, and audit trail as the rest of Carelytic. And it's included in every tier — no AI add-on, no per-seat AI fee, no per-claim AI charge. What you sign up for is what you get.
See everything Sage does → · Book a demo to watch it run on your own workflows.