ConnectHealth: AI EHR Integration Platform (FHIR & HL7)

ConnectHealth v0.4.0 Release Notes

This release brings a richer Mapping Engine, with FHIR Bundle output, built-in value conversions, several-to-one field mapping and C-CDA as both source and target, a new Run Inspector that shows every workflow execution step by step, and built-in guided tutorials across every page and 80 workflow nodes. It adds CDS Hooks for Epic connections, workflows that AI assistants can call as tools, a connector for EHR APIs beyond FHIR (starting with Open Dental), and a C-CDA parsing node. ConnectHealth DirectLicense lets a deployment be licensed directly by ConnectHealth, alongside or instead of an AWS Marketplace subscription. Connection setup is now a guided, step-by-step flow, and account security adds two-factor sign-in and automatic logoff. The platform now runs on a shared in-memory data service (Redis) that this version's template adds to your stack.

Existing connections, workflows and saved mappings continue to work. Please read the Migration Notes before upgrading: this version requires an updated deployer user policy, and every user signs in again after the upgrade.

HIGHLIGHTS

A Richer Mapping Engine

• FHIR Bundle output -- select several FHIR resources as the target and the mapping produces one Bundle, with a condition available on each resource
• Built-in conversions -- dates and times (FHIR and HL7 formats), letter case, trimming, replacing values, number and true/false conversion, and list handling (split, join, pick, count, sort). Conversions run in the order they are added, with a live preview in the Field rules window
• Several sources into one field -- join up to ten source fields with a separator, or take the first non-empty value
• Readable HL7 v2 and X12 names -- every HL7 v2.1 to v2.8 segment and field, and every X12 005010 segment and element, is shown by name as well as by code
• Output templates -- the target sample becomes the output template, and any target field can be given a fixed value
• Editor refinements -- repeating items in the target tree are shown as a single list, template edits are preserved across save and reload, and closing the Field rules window asks before discarding changes

Run Inspector

• Every execution at a glance -- Logs now opens on a Runs view with one row per workflow execution: status, trigger, number of steps, duration and the step where it failed
• Filters and grouping -- filter runs, group them by workflow with typical durations, and let the list refresh automatically
• Step-by-step timeline -- open a run to see each step on a timeline. Errors are reported by type and code. Data values and patient information are never recorded
• The previous event stream remains available under Raw

Built-In Guided Tutorials

• Help on every page -- a help button in the header lists the guides for the current screen, with an option to turn off automatic guides
• Workflow editor onboarding -- a short tour introduces the workflow editor the first time it is opened
• Guides for 80 nodes -- every ConnectHealth node and the common built-in nodes have a step-by-step guide with example flows that can be added to the canvas in one click

C-CDA End to End

• C-CDA Parse node -- turns any C-CDA document, such as a CCD, discharge summary or referral note, into structured data or a FHIR Bundle in one step. Malformed documents are reported instead of producing empty output
• Map from C-CDA -- C-CDA is now a Mapping Engine source, with readable section names and automatic suggestions for patients, problems, allergies, medications, results, vitals, immunizations, procedures and encounters
• Map to C-CDA -- C-CDA is also an output format. A starter sample pre-fills the required document structure, and output follows the C-CDA element order
• The Mapper node accepts the C-CDA Parse node's output directly

CDS Hooks for Epic

• Decision support inside the EHR -- an Epic connection can answer CDS Hooks requests with cards produced by a ConnectHealth workflow
• Three hooks -- patient view, order select and order sign
• Two new workflow nodes -- CDS Hooks In receives the request and CDS Hooks Response returns the cards. If a workflow does not answer in time, the EHR receives an empty response rather than an error
• Verified requests -- every request from the EHR is checked against the EHR's published signing keys before a workflow runs
• Simple setup -- the connection page shows the discovery address to give the EHR, the hooks in use, and the EHR signing-key address to enter

Workflows as AI Tools

• Publish a workflow as a tool -- the new MCP Tool and MCP Response nodes let AI assistants and agents discover and call a deployed workflow over the Model Context Protocol
• Protected and audited -- calls require your API key, and every call is recorded in the audit trail

Beyond FHIR -- the EMR Connector

• A new EMR Connector node calls an EHR's own API for the data FHIR does not cover. Open Dental is the first EHR supported
• Endpoint catalog -- choose from a catalog of the EHR's endpoints instead of composing requests by hand
• Credentials from the connection -- the node uses the keys stored on the Open Dental connection you already set up
• Clear errors -- a not-found or rejected request from the EHR is reported as such, with guidance on how to resolve it

WORKFLOW BUILDER

More Workflow Improvements

• Shared AWS settings for queue and database nodes -- the SQS and DynamoDB nodes can now use the same AWS configuration node as the Athena node. Existing per-node settings continue to work
• Nodes arrive with real names -- a node dropped onto the canvas is named after what it does, so exported workflows and AI assistants can read each step
• Unsaved-changes warning -- leaving the workflow editor with undeployed changes asks whether to stay or discard
• The join node is now called Collect. Existing flows that use the previous name keep working
• AI providers -- the AI Model node offers Amazon Bedrock and Google Vertex AI, with Bedrock as the default. A rejected AI credential is now reported as an invalid credential for that provider
• The AI Model node accepts a temperature of zero

CONNECTIONS

• Step-by-step setup -- creating a connection is now a guided flow built for each EHR. It shows the values you give the EHR, such as callback, launch and key addresses, before asking for yours. Each step stays editable
• One clear settings page -- a published connection opens on a page showing all of its settings, read-only until you choose Edit
• Duplicate HL7 routing keys are reported when a draft is saved, instead of being saved silently
• CharmHealth persistent connection -- workflows can reach CharmHealth system to system using a long-lived connection. CharmHealth single sign-on now completes and returns to the application. CharmHealth persistent connections do not count toward your connection limit at this time

LICENSING AND CONNECTHEALTH DIRECTLICENSE

• ConnectHealth DirectLicense -- licenses purchased directly from ConnectHealth can be used on their own or alongside an AWS Marketplace subscription, and connection limits from both add up. Each license answer is cryptographically signed and checked before it counts. A License Sources card in Settings, Subscription shows the status of each source, and each connection shows which source it counts against
• Sandbox connections are free -- connections to an EHR vendor's published test environment no longer use a licensed connection slot, and may be created past your plan limit
• Accurate Marketplace counting -- only your ConnectHealth subscription counts toward your connection limit. Previously, other AWS Marketplace subscriptions in the same account could be counted
• A short outage of a license source does not disable connections. A 24-hour grace period applies

SECURITY

Two-Factor Sign-In and Automatic Logoff

• Two-factor authentication -- users set up an authenticator app from Settings and verify with a one-time code. Each user can then require a code at every sign-in
• Recovery codes -- ten one-time recovery codes are shown once at setup and can be downloaded. They are never stored in the browser
• Automatic logoff -- after 15 minutes without activity a countdown warning appears, then the session ends. This supports the HIPAA automatic-logoff safeguard. Signing out in one browser tab signs out every tab
• Active users stay signed in -- a session renews itself while someone is working, up to a 9-hour limit
• One active session per user -- signing in on a second device ends the earlier session. The change is recorded in the audit trail
• Fully audited -- two-factor setup, sign-in codes, recovery-code use, resets and logoffs are all recorded in the audit trail

Data Protection

• Credentials are kept out of workflow messages -- AI provider keys and service-account credentials no longer travel inside workflow messages, so they cannot appear in Debug output or error handlers. They are looked up only at the moment a call is made
• Fewer credential traces in logs -- EHR client-secret fragments are no longer written to application logs, and AI provider errors are logged without request headers or prompts
• Stronger key validation -- signing keys and private keys are validated cryptographically when a connection is saved, not only when it is tested. Weak, mismatched or malformed keys are rejected, and every rejection is recorded in the audit trail
• Credentials stay out of the browser -- connection secrets are never written to the browser's draft storage while a connection is being set up
• Encrypted chat storage -- new Helix chat history is encrypted with your stack's compliance key
• HL7 connector VPN -- the second VPN tunnel now uses the same encryption settings as the first, so a failover can no longer fall back to weaker defaults. A separate pre-shared key can now be set for the second tunnel
• Protected data service -- the new Redis service uses a generated password stored in AWS Secrets Manager, accepts connections only from ConnectHealth application tasks, and runs with no AWS permissions of its own

USAGE DIAGNOSTICS

• ConnectHealth now collects anonymous product usage information: which features and workflow node types are used, and how often. It never includes workflow names, data values, user identities or patient information. Your deployment is identified only by a random identifier
• Usage diagnostics are on by default. An administrator can switch them off at any time in Settings, General. The same page shows your Deployment ID, environment and version

BUG FIXES AND REFINEMENTS

• Audit Logs -- moving to the next page no longer repeats events from the previous page
• Application Logs -- pages no longer come back partly empty, and the log view no longer sticks on loading
• Connection test -- the refresh-token step of Test Connection works again
• Connections -- a new connection can be published without saving and reopening a draft. Test Connection is disabled while a saved connection is being edited, and Cancel restores the saved values without a reload
• Mapping import -- re-importing a workflow whose mappings already exist asks whether to overwrite or create a copy, instead of showing a misleading message. Each import is recorded once in the audit trail
• Mapping and workflow lists -- deleting the last item on a page no longer leaves an empty extra page, and counts stay correct after a delete
• Workflows -- created and updated dates now show their stored values
• Helix settings -- Google Vertex AI credentials are kept when settings are saved
• Sign-in -- a user with a pending password reset is always taken to the reset page first
• Dark mode -- delete and remove warnings are readable
• HL7 -- help text and connection copy no longer refer to deployment tooling

MIGRATION NOTES

• Action required: deployer user policy. Apply the user policy included with this version before upgrading the stack. It adds the permissions needed to create the Redis service and its private DNS name, and consolidates the previous five policies into four.
• Action required: Redis. Upgrade using the 0.4.0 template. It adds a Redis service, its private DNS name, a generated password in AWS Secrets Manager, and network rules. Four new parameters control it: Redis Image, Redis Task CPU (default 512), Redis Task Memory (default 1024 MB) and Redis Max Memory (default 650 MB; keep it near 60 to 65 percent of the task memory).