Six plays to scale trusted care.
Build on the edge relationship already in place. Each play maps to a Hims product or roadmap item documented publicly; Cloudflare is the proposed enabling layer, not a replacement for clinical systems or clinician judgment.
GAI Gateway + AI Security
One governed front door for every model
Standardize how Labs AI, MedMatch, and future agents reach Workers AI or third-party models.
- Provider routing, fallback, rate limits, and cost visibility
- DLP and prompt-injection detections as defense-in-depth
- Metadata-only logging and no patient-specific caching for PHI paths
AAgents SDK + Durable Objects
Stateful care agents that preserve continuity
Give each member or care journey durable state, resumable conversations, scheduled follow-up, and explicit handoff.
- Persistent sessions across app, web, and reconnects
- Human approval before clinical actions or EMR writeback
- Patient-isolated state with retention and deletion controls
RAI Search + Vectorize
Retrieval grounded in Hims clinical knowledge
Extend the curated-knowledge pattern behind Labs AI into provider, support, medication, and operations copilots.
- Hybrid retrieval with source citations and metadata filters
- Policy versioning, expiration, authorization, and abstention
- Curated evidence instead of open-web answers
WWorkflows + Queues
Durable orchestration beyond the chat window
Coordinate intake, routing, documentation, lab follow-up, refill exceptions, and pharmacy operations through recoverable steps.
- Retries, waits, approvals, and complete step history
- Deterministic escalation for urgent and high-risk cases
- Backpressure and dead-letter handling for operational work
MEmail + Voice + MCP + Browser Run
Meet members and teams where care happens
Make agents available through chat, voice, email, push, and approved operational tools without opening broad data paths.
- Minimal-PHI notifications with consent and preference controls
- Scoped MCP tools and short-lived user-delegated identity
- Human-assisted browser automation only where APIs do not exist
OAgent Tracing + Flagship + Sandbox
Ship agents like regulated software
Observe every model and tool step, test against clinician-created benchmarks, and roll out safely by cohort.
- Agent session tracing with PHI-redacted telemetry
- Shadow releases, feature flags, instant rollback, and kill switches
- Isolated code execution for evaluation and engineering agents
New from Agents Week
Current through Aug. 4, 2026
Cloudflare AgentsUnified agent sessions and tracing. Beta.
Python ↔ JavaScript RPCConnect ML services and application agents. Available.
@cloudflare/computerAgent runtime across isolates and containers. Early preview.
Billable Usage APIProgrammatic AI and platform FinOps. Live.
The clinical line stays bright.
Cloudflare security controls can help govern AI traffic, tools, and data, but they do not validate medical accuracy. Diagnosis, prescribing, dose changes, abnormal-lab interpretation, urgent triage disposition, and EMR writeback should remain behind validated deterministic rules and mandatory clinician approval.