Agent OS
Describe the work. Agent OS builds the agent.
Tell Console Agent what you need in plain English. It builds the agent, tests it, fixes what fails, and puts it to work on your EHR, with the same guardrails as Shift’s own agents.
A team member asks Console Agent in plain English: “Build an agent that works every new MRI referral: check the payer’s auth rules, submit the auth with what the payer wants attached, and hand it to our auth team if there’s no decision in 2 days.” Console Agent builds the agent, writes 24 tests, fixes the one that fails, and the team publishes it. The published MRI Referral Agent then works a referral for Nina A. · Lumbar MRI: Mon 8:14, Fax split, matched to Nina’s chart. Mon 8:16, Active. This plan requires prior auth for MRI. Mon 9:02, This payer wants 6 weeks of PT notes. Attached from the chart, auth submitted. Tue 9:00, Portal: still pending. Next check Wed 9:00. Wed 9:00, Approved. Procedure scheduled Thu 7:30am. Referral to scheduled procedure in 2 days. Escalation never needed.
Use Shift’s agents
Nine agents, ready on day one.
Patient Access, RCM and Contact Center agents, configured to your payers, your SOPs and your EHR by our team.
See all agents →Build on Agent OS
License the platform. Build your own, or have us build it.
Your engineers build, test and run agents on the same infrastructure as Shift’s own. Or our implementation team builds custom agents for you.
Talk to us about licensing →The platform
Everything an agent needs, in one system.
From the channel a patient uses to the system of record, every layer is built in, and governed the same way.
Cases
Work that outlasts a conversation.
When an agent needs to wait on a payer, a fax or a patient, it opens a case and comes back on your schedule.
- Statuses: active, waiting, escalated, resolved
- Check-in ladders with the steps, gaps and channels you approve
- A maximum lifetime, then a handoff to the person or group you name
- A reply from the patient resumes the case immediately
Example
Auth submitted Mon → check the portal Tue 9:00 → no decision by Wed noon → auth team, with history.
Cases
Built for work that takes days, not one conversation.
A referral waits on a fax. An auth waits on a payer. A balance waits on a patient. When work can’t finish in one go, Agent OS opens a case, checks back on the schedule you set, tries another channel, and hands off to a person, with the full history, before the deadline.
- Active
- Waiting
- Escalated
- Resolved
A new message from the patient cancels the pending check-in, so the case picks up from the reply instead of sending an outdated reminder.
Cases & Memory →Check-in ladder · Patient Balances
Statement sent. No response.
- +30 min Send a gentle reminder by text, with the secure pay link.
- +4 hrs Try another channel: email the statement.
- +3 days Call once, during the patient’s preferred hours. Memory
- +7 days Escalate to the billing team, with every attempt attached.
Your rules: the steps, the gaps, the channels, the maximum case lifetime, and who gets the handoff.
Build · Test · Publish · Observe
Ship agents the way you ship software.
Nothing reaches a patient or a payer until it’s tested, versioned and published, and you can always see what it did.
Instructions
Model
Skills
Tools
Knowledge & memory
Cases
New in this draft
Tools & integrations
Connects the way each system allows.
Every action is a Tool with one contract, whatever is underneath. Reads run free. Writes are gated and read back to confirm they posted.
APIs
EHR/PM, documents, payments: wherever a real API exists.
EDI
270/271 eligibility, 837 claims, 835 remits, in the standard format.
Browser Agents
Payer portals and EHR screens no API reaches, including two-factor logins.
Databases
Direct queries against your reporting database or warehouse.
Documents & fax
Classify, split and extract from any PDF, flagged for review when unsure.
Triggered by an event, a schedule, a webhook, or a work queue that skips what it’s already done and retries what failed. See integrations →
Governance
You set how much it does on its own.
Tier 1
Assist
Agent surfaces, human applies.
Tier 2
Supervised
Agent acts, human reviews everything.
Tier 3 · recommended
Exception-based
Acts on high confidence, routes the rest.
Tier 4
Autonomous
Acts on nearly everything, human spot-checks.
Per agent, per task. Every write confidence-scored, gated and read back. Full audit trail with per-action rationale. BAA with every customer. Security & compliance →
The library
Start from nine agents, not a blank page.
Use them as they are, extend them, or borrow their parts. They share one payer-rules knowledge base, one coding engine, one responsibility estimator and one memory.
Patient Access RCM Contact Center
See all agents →See Agent OS run on your own EHR.
Bring a workflow your team runs by hand. We’ll show you it built, tested and running.