Healthcare Administrative Automation
K

Keragon

Keragon is an automation platform, sold as HIPAA compliant, where practices and digital health companies build workflows and AI agents across the software they already run. Workflows are deterministic, no code automations for scheduling, intake, billing and record sync that run the same way every time. Agents, still in early access, take on work that needs judgment: intake, reminders, no show recovery, answering questions, triaging messages and summarizing charts, with role templates for a receptionist, billing and RCM, scheduling, insurance verification, engagement and intake. The AI Editor drafts a workflow from a plain language request, and Keragon MCP, in beta, lets a customer's own ChatGPT, Claude or voice AI act on connected tools under a business associate agreement. The platform connects more than 400 tools and 50 EHRs, including athenahealth, AdvancedMD, Tebra, DrChrono, Healthie, Elation and eClinicalWorks, through official APIs only.

More than 1,000 healthcare companies use it, running about 50,000 automations a day and 20 million to date. The Autism Center of Illinois cut onboarding by three days and saves 10 hours of admin a week, the Institute of Contextual Health saves 10 to 15 hours a week, and Timber Creek Counseling saves 10 minutes per new client. Conno Christou leads the New York company, which he cofounded in 2022, and Keragon was a launch partner for Gemini for healthcare. Plans start at $149 a month, or $99 billed annually, with a BAA on every paid plan.

AI Health Index verifiedNovember 19, 2026
Compare Keragon with other vendors
Founded
2022
Headquarters
New York, New York
Categories
healthcare-admin-automation, rcm-and-prior-auth
Indexed Products
Keragon Agents, Keragon Workflows, Keragon AI Editor, Keragon MCP
Buyer Segments
Independent and multisite practices, Digital health operators, Healthcare operations teams, Enterprise integration teams
Assessment

Capability Axes

An AI Health Index grade measures what a buyer can verify from public sources on the date shown. It is not a rating of how good the product is. A vendor can build an excellent system and grade low on an axis because it publishes nothing an outsider can check. How grades read

AI Capability
BB on AI CentralityThe model is the engine of a core module. The platform carries other value, but this capability does not exist without it.
Vendor Published

Two engines run side by side. Workflows are deterministic: a practice wires triggers and connector actions into fixed sequences for forms, appointments, billing and record updates, and they run without a model. Agents and the AI Editor are where models come in. An agent uses a built in model to classify patient replies, prepare administrative messages, summarize charts or decide which workflow should handle a case, and the editor turns a prompt into triggers, actions and logic.

Both paths share the same 400 connectors, permissions and audit trail. The model is the engine of the agent and editor functions, while the workflow engine behind most of the 50,000 automations a day stands on its own.

AA on Autonomy and Oversight ModelWhat the system may do and what it may not do are both published, with escalation thresholds, override paths and the conditions that route a case to a person.
Vendor Published

A designer chooses which systems and which individual actions an agent may use. Connectors default to read access; a write action has to be switched on, and the designer can put a human checkpoint in front of it. The AI Editor offers a planning mode that lays out steps for confirmation before it builds, or it can draft directly, and an administrator approves publication of the resulting workflow.

Test runs show results before activation, failed runs raise alerts, and during a live run case specific checkpoints pause the action until a person approves. Afterward the practice can read the action history and revoke a single permission. Agents remain in early access, so these controls govern a feature still settling into production use.

BB on Model and Technology TransparencyThe approach or the suppliers are named without the version and update discipline behind them. Naming a supplier is the entry to this band both here and on Model Supply Chain Disclosure, which ask different questions of the same disclosure: who receives the data, and what produces the output.
Vendor Published

The public subprocessor list names Anthropic's Claude API, configured with zero data retention and HIPAA enabled, as the model behind the native agents, and the built in agents run without a customer key. Enterprise customers can bring their own model provider. Keragon MCP connects outside assistants, whether ChatGPT, Claude, a voice AI or Gemini for healthcare, to the actions a practice permits in its connected software, and the Gemini connection launched with Keragon as a partner.

The AI Editor turns a request into a sequence of triggers and actions. That identifies the supplier and the integration paths. Which Claude model and version each feature uses, and when it changes, are not stated for the editor or the agents.

BB on Model Supply Chain DisclosureSubstantial partial disclosure, or a chain that is structurally short: an in house build, a cleared model that cannot be quietly swapped, or a deployment where the transfer does not occur at all. Naming the model provider exits the band below into this one; the axis rises from here on the completeness of the party list and on the terms that govern data once it arrives.
Vendor Published

The trust center's public subprocessor list enumerates 16 parties: Amazon Web Services for hosting, Anthropic for the Claude API with zero data retention and HIPAA enabled, and business services from Stripe and Clerk to Zendesk, Vanta, Notion and Fireflies. The company states that neither it nor its model provider trains on customer data, and the data processing addendum sets processor obligations.

Parties a customer adds sit outside that list: an Enterprise customer's own model provider, an external voice service for calls, or a ChatGPT or Gemini session acting through MCP, each under the customer's own terms. For the native agents the chain is named with the model's retention and training terms, while routing for those customer chosen paths depends on each deployment.

CC on Clinical and Operational EvidenceNamed customers, or vendor reported percentages with no method, denominator or reference standard. Scale of use is recorded here and is not treated as evidence of benefit.
Vendor Published

Three customers carry the case studies. The Autism Center of Illinois cut onboarding by three days and saves 10 hours of administrative work a week, according to founder and CEO Dr. Sarah Lewis. The Institute of Contextual Health saves 10 to 15 hours a week, half a staff position, on fax, task and RingCentral messaging workflows, and Timber Creek Counseling saves 10 minutes on each new client intake.

Women's Mental Health Specialists credits referral intake automation with five hours a week and a share of 15% revenue growth. Platform scale runs to 20 million workflow executions. Each result belongs to a deterministic workflow, and none isolates what the current agents add or measures an agent's decisions against staff.

AA on AI Safety and PHI StewardshipRetention windows, training use and de identification are all stated specifically enough to be contradicted. Safety engineering, meaning guardrails, hallucination mitigation and how a safety event is handled, does not gate this grade: no vendor in this index publishes it, so every note on this axis states that gap rather than leaving it implied.
Vendor Published

AI inference is kept apart from the records a customer chooses to store in an automation. Model calls run under zero data retention at Anthropic, and the company says neither it nor its model provider trains on customer data. The data processing addendum covers anonymized data and data masking options. Encrypted daily backups are kept 14 days, execution logs 7 days on standard plans, and data is deleted when a customer leaves.

Encryption is AES 256 at rest and TLS 1.2 or higher in transit, and integrations use official APIs only, with no screen scraping. Test runs, approval gates and scoped permissions limit action errors, and incident response covers confirmed data breaches, though no procedure is set out for an agent that takes a wrong action on a record.

Regulatory and Compliance
BB on HIPAA and BAA PostureBusiness associate status is stated and supported by a substantive privacy document, with the agreement or its scope not fully published. For a vendor outside the United States, an equivalent regime documented to this depth grades here. A vendor whose product does not process protected health information grades here too when it says so plainly and explains the scope, with a service privacy document behind it: stating a position a buyer can rely on is the posture this axis grades, and the band above is closed to it because there is no agreement to publish.
Vendor Published

A standard business associate agreement comes with every paid plan, from Starter up, and an eligible organization can open and sign it inside the dashboard, even during the trial. Enterprise customers can negotiate custom terms, which carry a one time legal review fee. Data flowing through the native agents and through Keragon MCP falls under the same agreement.

The public data processing addendum requires authorized subprocessors to meet protective terms, but that is a separate document, and the BAA itself sits behind an access request in the trust center. The plan and signing path are clear before purchase, while the agreement's own treatment of subcontractors becomes readable only once the form is opened in the dashboard or the trust center.

BB on Security Certifications and Trust CenterA recognized certification is named in the vendor own material without the artifact, or with a scope or renewal question the buyer has to raise. A recognized certification here is an audit of how the vendor operates security, such as SOC 2, HITRUST or ISO/IEC 27001. A certification has a scope and a clock, and both are part of this grade. A certification the vendor attaches to its data center, hosting facility or cloud provider, rather than presenting as its own, is the host's and does not count here.
Vendor Published

Prescient Security audited Keragon under SOC 2 Type II, and Vanta monitors controls continuously. The trust center lists the 2026 to 2027 SOC 2 report and a separate HIPAA assessment behind an access request, alongside public control summaries for encryption, data retention, deletion on exit and production database authentication. The company runs regular penetration tests and a vulnerability disclosure program, offers SSO on Enterprise and commits to a 99.9% uptime SLA.

Credentials are centrally managed, and connector access is scoped to read or write. The auditor and report type are named. The report itself, its period and the systems in scope come through the access request rather than self service.

CC on FDA and Regulatory StatusNo device claim is made and the product is scoped accordingly. Most administrative and operational products sit here and are not penalized for it, because this axis grades the appropriateness of the positioning rather than possession of a clearance. A stated limit on use, such as research use only or a clause that the product gives no medical advice, beside marketing that sells the product for clinical decisions also sits here, with both quoted, because the buyer has to ask which one governs.
Vendor Published

The workflows move administrative information between systems: referrals, appointments, intake forms, insurance checks and billing. Agent actions can create a note or update a record inside a practice's existing software, and the patient communication tools pick a template, draft administrative text or route a reply, while clinical questions can stay with the practice's staff. The product is scoped to healthcare operations and system integration.

No medical device claim is made for the agents, and none of the role templates, from the receptionist to insurance verification, offers diagnosis or treatment advice, so the administrative positioning matches what is sold.

BB on AI Governance and Bias DisclosureA governance framework with named process behind it, such as certification to an artificial intelligence management standard, or material written for a customer own review committee to evaluate the product with. A framework covering the vendor whole portfolio counts when its scope reaches this product; what it cannot supply is the product specific evaluation the band above asks for. Published results within the one group where the performance of this product is most at risk also sit here: evaluation on the question this axis asks, short of the comparison across groups the band above requires. A medical device clearance or certification, and the quality system behind it, does not reach this band on its own; it is graded under regulatory status. A regulator reviewed plan governing how the model may change does reach it.
Vendor Published

An organization chooses each agent's permitted actions, can require approvals and tests outputs before activation. Failed runs raise alerts, and the activity log ties each action to the workflow and to the user or agent that started it, with always on audit logs on Enterprise. The company says it maintains evaluations and safety rules for agents and runs compliance monitoring through Vanta. That gives a practice concrete controls to review an agent and reconstruct its operating history.

Evaluation by population is what is missing: no results show how accurately agents classify messages from patients writing in different languages or at different reading levels, and none compare error rates between a dental group, a counseling practice and a virtual care company.

CC on AI Liability and RecourseMechanisms exist that let someone challenge an output, such as audit trails, source traceability or review before commit, with nothing standing behind the output and no route for the harmed party.
Vendor Published

A workflow owner can inspect each run log and the agent's action history, the editor keeps versions for rollback, and the customer can require review before a sensitive action changes a connected record. Those let a practice investigate an error and undo it. The service terms disclaim warranties about accuracy or the correction of errors and cap liability at the greater of $100 or fees paid in the three months before the customer first asserts a claim, subject to applicable law. The 99.9% uptime SLA covers availability. No remedy or correction obligation attaches to an agent that books the wrong slot, sends the wrong message or writes to the wrong chart.

Integration and Deployment
AA on EHR and Interoperability DepthNamed bidirectional integrations with major record systems, verifiable in marketplace listings or integration documentation, with evidence the connection runs in production.
Vendor Published

The eClinicalWorks connector supports looking up patients and coverage, creating or updating patient records and writing notes and documents, and it is enabled through eClinicalWorks production activation. Deep coverage extends to AdvancedMD, Tebra, athenahealth, DrChrono and Healthie, with Elation, ModMed, Open Dental, Dentrix Ascend and IntakeQ among the 50 EHRs supported, all through official APIs rather than robotic automation.

Jack Pine Media used DrChrono in production to create patient records for a clinic client and reported a more limited scheduling connection. An organization picks which connected actions an agent may take, read by default. Action sets differ by system, so the write path depends on both the EHR and the customer's setup.

BB on Deployment Model and Data ResidencyOptions and residency are stated with isolation or the processing path left open.
Vendor Published

The hosted platform runs on Amazon Web Services in US East, with redundant availability zones, encrypted backups and a standby database. Enterprise adds multi region hosting and VPC peering, along with log streaming to the customer's own endpoint and full text log search. Administrators set which staff can view, build or manage workflows and which system actions an agent may call. That states the default location and the enterprise options. Where Anthropic processes a model call is not given, the multi region option names no regions, and isolation between customer organizations is described only through access controls.

Commercial
AA on Commercial TransparencyPublished tiers with figures, a stated unit of charge, and a route to start without a sales conversation.
Vendor Published

Pricing is public and self service. Starter costs $149 a month or $99 a month billed annually, Professional $399 or $269, and Scale Up $1,499 or $995; Enterprise is custom and annual only. Starter includes 200 workflow runs, 5,000 AI credits and three published agents or workflows, Professional 2,000 runs, 15,000 credits and 15, and Scale Up 20,000 runs, 60,000 credits and 100.

Extra runs cost $0.15, $0.07 and $0.02 by tier, AI credits run $0.0075 at the standard rate with discounts on higher plans, and Starter agents pause at their credit cap. Users are unlimited, a 14 day trial needs no card, and MCP is free in beta at 100 tool calls a day.

BB on Setting and Specialty CoverageCoverage is named with validation behind part of it.
Vendor Published

The buyers are independent practices, multisite groups and virtual care or digital health companies, and the customer logos run from MB2 Dental to counseling and integrative practices. The named cases cover autism services (Autism Center of Illinois), behavioral health counseling (Timber Creek Counseling), chronic pain care with messaging and fax automation (Institute of Contextual Health) and virtual women's mental health (Women's Mental Health Specialists).

Templates and examples reach into primary care, dentistry and wellness. Those cases support the operational settings for workflows. The agents, still in early access, have no comparable results in any specialty yet.

Alternatives

Alternatives to Keragon

Other products that do the same kind of work as Keragon, and what each one does.

GenHealth.ai supplies AI agents that perform healthcare administrative work inside existing applications. The current offer spans intake, eligibility, prior authorization, resupply, billing and denial follow up through web interfaces, fax and telephone.

Flexbone, based in Atlanta, builds AI agents for administrative operations, primarily in healthcare; its privacy policy describes it as a healthcare automation company working with providers and payors.

Emitrr is a patient communication platform for independent practices that puts texting, phones, faxing, online forms, reviews and scheduling in one system, with AI agents that take calls and messages the front desk cannot.

Phreesia sells patient intake and access software to health systems, enterprise groups and independent practices. Founded in 2005 and listed on the New York Stock Exchange, it says more than 4,700 healthcare organizations use it and that it supports 1 in 6 US patient visits.

HealOS, the company formerly called ScribeHealth AI and operated by Better Financial Corporation, sells six AI agents for clinics.

Develop Health automates medication benefits verification and prior authorization for prescribers, pharmacies and other access teams.

Commercial

Pricing

Vendor-published figures are labeled as such. Figures labeled “Estimated” are derived from third-party sources and have not been confirmed by the vendor.

Entry Price Pricing Basis BAA Tier Implementation Source
$99/month billed annually
$99 baseline
Starter is $149 monthly or $99 a month billed annually ($1,188 a year). Professional is $399 or $269 ($3,228), and Scale Up is $1,499 or $995 ($11,940). Enterprise is custom and annual only. Each plan has monthly allowances for workflow runs and AI credits. A standard BAA comes with every paid plan and can be signed in the dashboard during the trial. Enterprise terms are negotiable, with a one time legal review fee. A 14 day trial needs no credit card. Customers build workflows themselves or hire specialists through the marketplace, and Enterprise includes an assigned workflow engineer. Vendor Published

Starter includes 200 runs and 5,000 AI credits, Professional 2,000 runs and 15,000 credits, and Scale Up 20,000 runs and 60,000 credits. Extra runs cost $0.15, $0.07 or $0.02 by plan, and Starter agents pause at the credit cap. MCP is free during beta at 100 tool calls a day.