MediBOT Agents runs inside your own AWS, connected to eClinicalWorks, claims and documents. Every team uses it right inside Microsoft Teams. And it does not wait to be asked.
Automation runs when it is triggered and follows fixed rules. Agents run continuously, reason over evidence, and act with judgment. The agent's user is not a person. It is the work itself.
Every automation that stops creates a queue item and a person who has to triage it. The agent reads the exception and the retry history, finds the common cause across items instead of retrying one at a time, and escalates only what needs a human.
Which workflows staff route around. Which reports stopped mattering. Which manual work repeats often enough that it should be automated next.
When a payer changes a portal or a mapping goes stale, the agent proposes the fix. A person approves it. Then it ships.
No scheduler sending billing reminders. It monitors every claim, understands aging, cross references the schedule, flags what needs attention and drafts the follow up, before the billing manager opens their inbox.
For eighteen months we built deterministic automations that log into payer portals, read result documents, and clean problem lists. They still run. But they are no longer the product. They are the tools the agent uses.
Eligibility verification, lab result entry, problem list cleanup — each does a fixed job when it is told to. The agent decides when to call each one, what to pass it, and what to do when it stops.
The agent sees across eligibility, claims, scheduling, clinical results and documents. It coordinates them as one system instead of letting each automation run in isolation.
When an automation fails, the agent groups exceptions by root cause, proposes a fix, and escalates only what needs a human. It refuses to guess and it never writes data it cannot validate.
The agent does not stop at a summary. When it sees something that needs a human, it opens the right work item in the EHR, routes it to the right staff, and gives them context. The loop closes inside your system, not in a chat thread.
A new lab, imaging or procedure result comes in. The agent reads the value, compares it to the order and the patient's history, and opens a telephone encounter for the care team when follow-up is indicated. Nothing falls between the result and the callback.
The agent watches the billing dashboard the same way a supervisor would. When a metric turns red — denials spiking, a payer pattern, accounts aging past a threshold — it creates an action alert with the account list, the reason, and the rework step, assigned to the right biller.
Every alert lands as real work in eClinicalWorks: a task, an encounter, a queue item. Staff do not copy data out of a chat. The agent writes the structured note, sets the priority, and tracks whether it was completed. That is the difference between a copilot and a colleague.
| Dimension | ✕ Copilot / ChatGPT | ✓ MediBOT Agents |
|---|---|---|
| Where your PHI lives | Multi tenant vendor cloud. | Inside your own AWS account. It never leaves your tenant. |
| What it can actually see | Outlook, SharePoint, Office docs. Not your EHR. | eClinicalWorks, 837/835 claims, denials, UDS extracts, credentialing, SOPs. |
| How well it knows the work | Generic model. No health center context. | Enterprise Intelligence Framework: eCW data models, UDS Table 6B logic, PR and mainland Medicaid rules, 340B economics. |
| Architecture | One chatbot with plugins. | Multiple specialized agents plus an orchestrator, supervising automations that already run. |
| Does it act on its own | Only when prompted. | Monitors, detects, decides, drafts. Proposes changes for human approval. |
| Inference economics | Frontier model on every query. Per seat fees forever. | Smaller models for routine work, frontier reasoning only where it pays. No per seat licensing. |
A generic model answers from the internet. MediBOT Agents answers from your health center: your vision and mission, your human-first values, your policies, your payer mix, your provider schedules, your referral relationships. Every conversation and every action carries that context with it.
The agent knows why the organization exists. When it drafts a response, prioritizes a worklist, or proposes a workflow change, it does so in line with the stated mission and values.
Standard operating procedures, compliance rules, prior authorization guidelines and payer specific requirements are embedded and referenced. The agent does not guess at policy. It checks it.
Provider schedules, panel assignments, site hours, referral networks and credentialing status are kept current. The agent uses them to route tasks to the right person at the right time.
The agent remembers prior questions, approved mappings and resolved exceptions. It does not treat every prompt as a blank slate. It learns the health center's language and applies it.
Every MediBOT Agent gets a mailbox inside your tenant. Staff can email it directly, forward a payer notice, or cc it on a task. It reads the message, understands the intent, and opens the right work item in your EHR — the same way it would answer in Teams.
A biller can forward a denial. A care manager can send a lab result. A front desk lead can assign a task by email. The agent sees it in its tenant mailbox and treats it as a request.
It parses PDFs, images, and message bodies, matches them to the right patient or claim, and extracts what matters. No special portal. No new login. Just an email address your staff already know how to use.
The agent does not reply with a summary and stop. It opens the telephone encounter, creates the billing action, or queues the automation — the same closed loop it runs from chat.
Many health centers still receive faxes and scanned referrals into a network folder. We treat that folder the same way we treat a lab interface: read the document, understand what it is, match it to the right patient and provider, and open the correct work item in eClinicalWorks. No new portal. No manual sorting.
The agent watches the shared scan folder, fax inbox, or document queue. PDFs and images are pulled in as soon as they land, with full audit logging of who sent what and when.
It identifies the document type — referral, prior auth, lab result, disability form, payer request — matches it to the patient chart, and routes it to the right department or staff member based on your rules.
Instead of a chat summary, the agent creates the referral task, flags the prior auth, attaches the document to the encounter, or opens a telephone follow-up. The same closed-loop judgment we use for lab and billing automation.
The agent is provisioned as a user in eClinicalWorks with its own credentials, role and queue. It does not sit outside the system pressing buttons. It works inside the EHR the same way a biller or care manager does: it opens tasks, writes structured notes, creates encounters and updates queues. The only difference is that it works them automatically, behind the scenes.
The agent gets a named user account in eCW, scoped to the roles and queues it needs. It follows the same permission model as staff: read where it should read, write where it is authorized, and nothing else. Auditors see exactly who did what.
Telephone encounters, billing action alerts, document tasks and exception queue items are assigned to the MediBOT user and worked in priority order. Staff see the same tasks, notes and statuses they are used to.
Because the work happens inside eClinicalWorks, front desk, billing and clinical teams do not need another dashboard. They interact with the agent in Teams or by email, and the output lands where they already look.
Not every question needs a frontier model. Most daily work — a denial summary, a schedule brief, a compliance check — is handled by local and smaller models running inside your AWS. No token charges. No external API calls. No per seat fees. When the work needs deeper reasoning, a router agent sends only that request to a frontier model, with a budget cap enforced inside your tenant.
Free and open source models run on your own infrastructure for routine queries, structured extractions and report formatting. Fast, private and predictable.
Multi step financial models, ambiguous exception clusters and policy drafts are routed to frontier reasoning models only when the task actually calls for it. The system decides based on the work, not the user.
Daily spend limits, per query ceilings and model specific budgets are set in your AWS account. If a cap is hit, the agent queues the work and tells you why.
Every incoming request is classified by intent, sensitivity and complexity. Simple lookups and formatted reports stay local. Reasoning, synthesis and ambiguous exceptions are flagged for frontier models.
A registry holds the local and frontier models the system can call, with cost per token, context window and capability tags. The router picks the cheapest model that can reliably do the job.
Common context — SOPs, payer rules, provider schedules, prior answers — is embedded once and reused. The same context is not re-embedded on every question, which cuts token spend and latency.
Before any action touches the EHR or a payer portal, a deterministic rule checks the output. The agent can draft, propose and queue. It cannot write unvalidated data.
This is what it looks like when MediBOT acts inside eClinicalWorks. It logs in as a staff user, navigates to the right module, enters structured data, and saves the record. No mystery API. No black box. The same clicks, the same fields, the same audit trail. These are real screen recordings of the agent performing work your staff used to do by hand.
The agent reads tomorrow's schedule, opens each patient chart, checks the payer portals, and writes the verified result back. Assertus, IMC, MCS, PSM, Triple-S. Consolidated report ready at 7:20 a.m.
The agent opens new result documents inside eCW, extracts the values, validates them against an accuracy rule, and enters them into the matching order as structured data. Missing fields are flagged, never guessed.
The agent opens each appointment, locates the patient by account number, applies the approved code and status mapping, and saves the record. One traceable queue item at a time.
Automations in production since January 2025, running against eClinicalWorks and five payer portals. Agent capability in pilot.
Analytics modernization delivered. Agent deployment in pilot, inside their own AWS environment.
You ask for a monthly clinical report. It does no heavy lifting. It breaks the request into a plan: fetch the data, format the report, publish it.
Takes the plan and knows exactly which digital employees are clocked in right now. Assigns each task to the right one.
Specialists. A data analyst. A report writer. A pipeline engineer. Each does real work and knows only its own domain.
Employees never touch the database. They ask a gatekeeper that fetches exactly what was requested. Even a confused AI cannot run a destructive command or take data it was not given.
| Your AWS account | Every deployment runs inside your own tenant. Your BAA with AWS. Your KMS keys. Your VPC. Your audit logs. We build it and we manage it. |
| Your data never trains a shared model | Model isolation is guaranteed at the platform level. Prompts, documents and outputs are not used to improve anyone's foundation model. |
| Credentials never sit in an agent | The resource servers hold the credential. A compromised or hallucinating agent cannot reach your database on its own. |
| Read only, by design | The warehouse connector is SELECT only with no schema rights. On premise databases are read from a replica, never a primary. |
| Row level security in the API | Billing sees billing. Clinical sees clinical. Enforced when the report runs, by the API, never by the language model. |
| Every call logged | Each tool call and each report access written with the requesting identity through Teams SSO. Your HIPAA access trail, on demand. |
| Changes are proposed, not applied | The agent that touches infrastructure drafts the change. A person approves before anything runs. Least privilege is not enough when the privilege is deploy. |
Every department, every leader, one interface. Nothing to file, nobody to wait on. Ask in Microsoft Teams, get the answer from your own data.
“Show top 10 denial reasons last month by payer, with dollars at risk.”
“List diabetic patients with A1c over 9 in the last 12 months, by provider panel.”
“Brief me on today's schedule: insurance flags, missing forms, new patients.”
“Should we hire another physician or an NP next quarter? Pull panel size, productivity, and payer mix.”
“What failed in last night's eligibility run, and what do they have in common?”
“Which providers have credentials or DEA licenses expiring in the next 90 days?”
One real workflow proven end to end before anything scales.
Your AWS environment and model access. Security, IAM and HIPAA controls. eClinicalWorks data model mapped. Staff access through Teams.
Retrieval pipeline built. EHR and claims data ingested. Policies and SOPs loaded. Department use cases configured. Accuracy tested with your own team, on your own numbers.
Training across every department. Reporting and automation enabled. Documentation handed over. Thirty days of monitoring after launch.
We pick the workflow with the clearest return, prove it end to end in your environment, and expand from something that already works. Thirty minutes, and we come prepared.
We bring community health center domain depth, eClinicalWorks expertise, a product in production and named references island and mainland. You bring reach, scale or cloud depth. Deployments run in the customer's own AWS account, so the consumption lands where you can co sell it.