AI agents
AI-first means the agents live in the modules
Not a chat window bolted onto a record from 2009. MIYO's agents run inside the workflow, on data already in the chart, and hand their output to a person for review before anything becomes clinical or financial fact.
What makes these agents different
No new data entry
Agents work from what your team already records — sessions, assessments, schedules, and claims. There is no second system to feed.
No new workflow
Output appears where the work already happens: the note, the claim, the calendar, the client's chart. Nobody learns a new tool.
No unattended action
Every clinical or financial output is reviewed and released by a person. The agent prepares; the practice decides.
The six agents
What each one takes off your desk
Documentation agent
Runs in Module 03
Drafts the progress note from the session data in the clinician's chosen format, proposes diagnosis and procedure codes, and logs progress against the active treatment plan goals.
A licensed clinician reads, edits, and signs every note. The agent cannot sign, and every draft is flagged as AI-assisted in the audit log.
Claims agent
Runs in Module 10
Turns the signed note into a coded, scrubbed claim, validates it against payer-specific rules, submits it, and drafts an appeal with supporting documentation when one is denied.
Billing staff approve the claim batch before submission and every appeal before it is sent. Nothing leaves the practice unreviewed.
Scheduling agent
Runs in Module 06
Matches clients to clinicians on availability, specialty, language, and stated preference, forecasts no-shows, and fills cancellations from the waitlist.
Suggested matches are proposals. Front-desk staff confirm every booking, and clinicians keep full control of their own availability rules.
Reporting agent
Runs in Module 11
Assembles compliance, utilization, and outcome reports on request, and translates session and outcome data into the language payers and grant funders expect.
Reports cite the underlying records, so every figure can be traced back to the encounters that produced it before anything is submitted.
Risk flagging agent
Runs in Module 08
Watches attendance patterns, screener trajectories, portal engagement, and medication adherence, and alerts the care team when signals combine into elevated risk.
Flags are decision support for a clinician, never a diagnosis or an automated action. Clinical judgment governs what happens next.
Workflow agent
Runs in Module 01 & 09
Runs the multi-step administrative chores end to end: consent renewals, authorization requests, supervisor co-signature routing, and recall outreach.
Anything that touches a client or a payer queues for staff approval. The agent prepares the work; a person releases it.
Governance
How we handle your clinical data
Small practices carry the same regulatory obligations as large ones with a fraction of the compliance staff. These are the commitments that matter when AI touches a chart.
Your data is not training data
Client records and clinical content are not used to train foundation models. Model improvement runs on de-identified, aggregated signals only, and never leaves the compliance boundary.
Every AI action is logged
Drafts, suggestions, and flags are recorded in the audit trail with a timestamp, the model version, and the person who reviewed them.
Part 2 material stays segmented
Agents respect the same consent and segmentation rules as human users. A record an agent may not disclose is a record it does not surface.
Attribution stays with the clinician
An AI-assisted note is still the clinician's note. The attestation, the licence, and the clinical responsibility belong to the person who signs it.
Questions we get asked
Before you let AI near a chart
Does the AI ever sign a note?
No. The documentation agent produces a draft. A licensed clinician reviews, edits where needed, and signs. The signature — and the clinical responsibility that comes with it — is always a person's.
What happens if the draft is wrong?
You edit it, the same as any draft. Corrections are captured, and the reviewing clinician's edits inform how the agent drafts for that clinician going forward. Nothing enters the record unreviewed.
Do you record sessions?
Ambient documentation requires client consent, captured and versioned in the intake and consent module. Audio is processed to produce the draft and is handled under your retention policy — you set whether it is retained at all.
Is our data used to train models?
No. Clinical content is not used to train foundation models. Aggregate, de-identified signals inform network-level benchmarks, and you can opt out of those.
How does this work with 42 CFR Part 2?
Agents operate under the same consent and segmentation rules as human users. Part 2 material is not surfaced, summarized, or disclosed outside the scope of a current, specific consent.
What if we want to turn an agent off?
Each agent is enabled per practice and per role. Some practices run documentation and claims and leave risk flagging off until they're ready. That is a setting, not a contract change.