AI steps in automations
Categorize an intake note, pull fields from a document or draft a reply, as one step in any automation.
AI agents
AI that works through the same doors your team does.
Agents have no back door into the database. They call the same APIs as the web app, under the permissions of the person asking. They propose, a person decides, and every change is logged with a name on it.
You
Track patient callbacks and remind the on call nurse after 4 hours.
Proposed changesPreview
- + Table Callbacks linked to Patient
- + Fields Reason, Due at, Assigned nurse
- + Interface Callback queue
- + Automation Remind on call nurse after 4 hours
RefineApply
Builder agent
Describe the app. Review the change. Apply it.
The builder agent turns a request into a proposal: the tables, fields, pages and automations it would add. You see a preview before anything changes. It starts from vetted pack shapes, not guesses.
- Propose, preview and apply, with undo
- Confirmation required before changes that touch patient data
- Works under your permissions, never more
You
Track patient callbacks and remind the on call nurse after 4 hours.
Proposed changesPreview
- + Table Callbacks linked to Patient
- + Fields Reason, Due at, Assigned nurse
- + Interface Callback queue
- + Automation Remind on call nurse after 4 hours
RefineApply
Triage · Ask
How many high acuity callbacks were late this week, by day?
7 late callbacks this week. Tuesday had the most, during the evening shift change.
Show workPin to Triage queue
Query plan
{
"table": "Tasks",
"filter": {
"type": "Callback",
"acuity": "High",
"late": true
},
"groupBy": "day",
"aggregate": "count"
}Read only · runs as youAnalyst agent
Ask a question. See the work.
Questions become structured queries, never raw SQL. The agent reads only what you can read, shows the query it ran and lets you pin the chart to a page.
- Read only by design
- Counts, groups and trends across linked tables
- Answers you can check
Everywhere else
AI inside workflows and outside the app.
An MCP server per app
Point your own assistants at an app through the Model Context Protocol. Read only, scoped by API key.
Every action attributed
Agent changes land in the event log with the agent and the approving person both recorded.
Guardrails
One gateway between your apps and every model.
All AI traffic goes through the HealthTable model gateway. The rules live there, so no feature can skip them.
Covered providers only
Apps that hold patient data send prompts only to model providers under a BAA. The gateway enforces it.
Budgets and limits
Spend budgets and rate limits per app, with redacted logging of every call.
A person decides
Agents draft and propose. In clinical apps a person confirms before anything changes.
Show us the workflow. We will show you the app.
Bring a process your team runs today. We will walk through how it looks on HealthTable, what is ready now and what is still on the roadmap.