Part of Connect Data & Intelligence
Less searching. More time for the next step.
Relay uses AI where it reduces work and keeps people accountable: Spark answers questions and drafts within each user's permissions, rules surface the work that needs attention, and agents propose actions that a person approves.
What your team can accomplish
Support your team's judgment with context and useful drafts.
Who it serves
- Care coordinators
- Supervisors
- Administrators who govern AI use
How AI use is governed
Use your own AI connection
Every model call goes through your organization's connection; there is no shared platform key. Without a connection, features return rule-based answers.
Control what reaches the model
Direct identifiers—Social Security and health plan member numbers, dates of birth, phone numbers, emails—are always removed. Names and street addresses are removed unless your organization allows them.
Require consent for member data
Drafts and summaries that use a member's information require that member's consent to AI use.
Govern prompts
A prompt registry with versions, approval, activation and rollback.
Watch cost and usage
A cost ledger, organization budgets, per-user limits, telemetry and an audit trail.
Each AI action and its boundary
Every AI capability in Relay, one at a time.
| Action | What it does | Can it change data? | Approval and limits |
|---|---|---|---|
| Spark answers | Answers questions about your work using eight read-only tools, scoped to your organization and permissions | No | Up to four tool rounds; needs your AI connection |
| Spark actions | Creates a referral or a task, or assigns a coordinator | Yes | You confirm each action; permissions, consent checks and audit apply |
| Drafts and summaries | Huddle summaries, care-plan drafts, communication summaries, suggested text replies | No—drafts only | Member data only with AI-use consent; a suggested text never sends itself |
| AI Command Center briefing | Ranks stalled referrals, needs without referrals, overdue tasks, expiring eligibility and more | No | Rule-based, no model; limited to each coordinator's members |
| Referral recovery | Finds stalled referrals and creates recovery tasks | Creates tasks | Rule-based, no model |
| Outreach agent | Drafts a first-name-only text or call script for members due for contact | Sends only after a coordinator approves | Consent, quiet hours, opt-out and duplicate checks at send time |
| Managed agents | Thirteen agents review referrals, quality, outreach, finance and more, and propose actions | Only after approval | Off by default; your organization may opt in to six internal-only actions running automatically |
Model-based features use only your organization's AI connection and its budgets.
AI Assistance & Workflow Automation
Your questions, answered
How does our organization get started?
For model-based features, connect your organization's Anthropic account. Rule-based briefings work without an AI connection.
When do AI actions require approval?
AI actions that change records require a person's approval. Your organization can explicitly enable specific internal actions to run automatically.
How are names and addresses handled?
Your organization decides whether member names may reach the AI model, with rule-based name and address detection as an added safeguard.
How are risk and community scores calculated?
Risk and community scores are transparent, rule-weighted formulas your team can explain.
How it connects to the rest of Relay
- Coordinate Care
The briefing and Spark actions work on the same member records.
- Identity & consent
AI-use consent is part of the consent ledger.
- Security & privacy
Access controls and audit history.
See AI assistance in Relay
We'll show each AI action, what it can do, and where a person stays in control.
Request a demo