Metriport
Metriport pulls a patient's records from health information exchanges, record systems, pharmacies, labs and admission and discharge feeds, then matches, deduplicates and normalizes them into one longitudinal record. Its Harmonization Engine does that work behind the Medical API, which returns consolidated FHIR R4 data, alongside a warehouse feed in a unified schema, a dashboard, embeddable views and prebuilt apps for record systems such as Healthie, Canvas and athenaOne. The company reports 4.2 billion network requests a month, more than 750,000 connected providers and records for more than 340 million people, with the first structured record arriving in under 15 seconds at the 80th percentile. In September 2026 it was accepted as a Candidate QHIN under TEFCA and moved into preproduction testing for direct national exchange.
On top of the record sit AI Summaries, which condense a patient's history into a short brief shaped by an account prompt, Condition Suspecting and Recapture, which flags possible undocumented conditions for a clinician to confirm, and HEDIS care gap identification. A ChatGPT plugin lets authorized users search a stored record, open source documents and read a summary that already exists.
Customers include Amazon One Medical, Sollis Health, Color, Maven Clinic, WeightWatchers, Brightside Health, Thyme Care, Hoag and Canvas Medical. Sollis clinicians pull prior EKGs, operative reports, imaging and labs in seconds, and at knownwell automating record pulls saves an hour of chart preparation per patient inside athenaOne.
Founded in 2022 by Dima Goncharov and Colin Elsinga and based in San Francisco, Metriport raised $26 million in August 2026 in a round led by Matrix with ARTIS and Y Combinator, for $28.4 million in total. Its core code is open source. Production access requires a treatment purpose on behalf of a covered entity, and AI Summaries is a premium feature turned on by request.
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
Record exchange, matching, normalization and deduplication are the core of the product, and they deliver structured FHIR data to other systems whether or not any AI runs, at a reported 4.2 billion network requests a month. AI Summaries is a separate premium feature that condenses the record into a brief tailored to what a practice wants to read, shown in the dashboard, an embedding application, the consolidated API response or a generated record summary.
Condition suspecting and care gap analytics add a second analytic layer for value based care teams. The August 2026 round funds a push toward clinical answers at the point of care, but the exchange still works without the generated narrative.
AI Summaries runs automatically when a customer requests the consolidated record with the feature on, and the brief sits beside the source records in the dashboard and embedded apps. Condition Suspecting works differently: a clinician confirms or rejects each candidate before it reaches the problem list.
The ChatGPT plugin reads existing summaries rather than generating them, shows source links, leaves ordering and clinical decisions with staff, and starts a new network query only on an explicit request. These are distinct controls for distinct outputs, with the firmest one on the output that changes the chart. No signoff is required before someone reads or reuses an AI brief, and there is no confidence threshold at which summarization pauses for review.
The current open source implementation includes a summary path on Anthropic Claude 3.5 Sonnet and another on OpenAI gpt-oss-120b, and the code picks between them by account configuration. It also shows how portions of a record are combined into the brief and how a customer's account prompt shapes the result, so an engineer can read the prompt assembly rather than infer it from output. The ChatGPT plugin generates nothing itself and only reads summaries that already exist.
Which model produces a given customer's production summaries is not stated, the models' training data is not characterized, and there is no policy for telling a customer when the model on its account changes.
The open source code names Anthropic's Claude on Amazon Bedrock for the summary and sets Bedrock as the provider for a newer open weight model path, with an adapter for Baseten that the checked configuration does not route customer traffic to. The ChatGPT plugin is a separate route, running inside OpenAI's plugin framework for organizations with MCP access enabled. Because the code is public, the model providers in the summary path can be read directly rather than taken on assurance. A complete roster of processors is not given, and the retention and training terms that apply to a clinical payload on each configured path, the plugin included, are not stated.
Named customers report concrete operational results. At knownwell, automating record pulls and chart preparation inside athenaOne saves an hour per patient, and Sollis Health clinicians see prior EKGs, operative reports, imaging and labs within seconds, which its chief medical officer ties to keeping patients out of the emergency room. Canvas Medical uses AI Summaries in context, and RTHM uses them to work through large records for complex illness.
The network figures, 4.2 billion requests a month and the first structured record in under 15 seconds at the 80th percentile, describe how fast records arrive. None of these results sets a measurement period, checks summary correctness against a reference, or separates the AI paragraph's contribution from the record pull itself.
The privacy notice says the company does not sell personal information and keeps service data for the customer relationship and a period afterward. In the open source code, selected records pass through configured AI processing paths to build the brief. The ChatGPT plugin limits itself to searching, following source links and opening existing summaries, has no tool that deletes records and returns at most 100 resources per search.
No fixed retention period is given for summary inputs or generated text, and no deidentification step or training restriction is stated for every brief and plugin path. Automated safety tests for the brief and a procedure for telling a practice about a harmful summary are not described.
Metriport positions itself as a subcontractor that helps covered entities and their business associates reach health information, calls the platform HIPAA compliant and makes a business associate agreement available. Production use of the Medical API needs an eligible covered entity relationship and a valid treatment purpose, which keeps network queries inside treatment exchange. The website terms separate public site use from activity that HIPAA covers.
The agreement text is not public, and its scope for each AI summary configuration is not stated. Nor is it stated whether the same agreement covers an organization's use of the ChatGPT plugin, where record content moves into OpenAI's service.
Metriport holds SOC 2 Type II and HITRUST r2 and lists AES 256 encryption at rest and in transit, TLS 1.3, end to end audit logging with provenance, and role based access on least privilege. Its trust center shows a 2025 to 2026 SOC 2 practitioner summary alongside retention, encryption, continuity and information security summaries, each behind an access request. HITRUST r2 is the more demanding HITRUST assessment and is a common requirement in health system procurement. The audit scope is not visible without access, so it is unclear whether those controls extend to the ChatGPT plugin and to every inference provider configured for the brief.
The medical record brief organizes clinical information a practitioner already holds, Condition Suspecting proposes possible conditions against configured definitions, and care gap identification flags HEDIS quality measures. The ChatGPT plugin searches stored records and reads existing summaries, and ChatGPT answers the user's questions from those resources.
In each path the clinical team interprets the source and decides what to do, and a suspected condition stays unconfirmed until staff act on it. No FDA clearance is claimed for the summaries or the suspecting feature. The product is positioned as information review and decision support, with no stated reasoning for why its clinical outputs fall outside device rules.
The open source code shows how the summary is assembled, how the system picks a model path and how an account prompt shifts the brief toward a clinical audience, which lets an engineering team audit the logic directly. The value based care tools add HEDIS care gaps and condition review, each working from configured definitions.
What is missing is oversight around releases: there is no approval process for changes to the summary, no fairness evaluation of summaries, and no comparison of omitted or wrong details across specialties or patient groups. No third party assessment of the summary is named either, so the trail a health system can follow stops at the code itself rather than at measured behavior in production.
Clinicians can open the original records in the workspace, and because the brief sits beside the source documents, a claim in the summary can be checked against the record it came from. The condition feature keeps a candidate unconfirmed until staff accept or reject it. The ChatGPT plugin returns source links for the resources it finds, reads only summaries that already exist and has no tool that deletes records.
The website terms give no remedy for an incorrect AI brief under a production agreement, and the privacy notice routes personal data requests without a set time to correct a clinical narrative. Compensation for a wrong summary, an outcome guarantee and a way for an affected patient to challenge a summary directly are not offered.
Canvas Medical uses Metriport to bring outside records into its chart and write Canvas updates back to connected systems, and an athenahealth Marketplace app renders summaries inside athenaOne, with one writeback function in beta. Healthie embeds the app, Qventus is a customer on the health IT side, and other named record systems include Epic, Elation and Practice Fusion.
The Medical API returns consolidated FHIR R4 data and can pass narrative documents in other formats, and messaging covers referrals, clinical data exchange and electronic case reports. In September 2026 the company was accepted as a Candidate QHIN under TEFCA and moved into preproduction testing, the step before direct national exchange. The actions available still vary by record system.
Metriport runs as a managed service reached through the API, embedded views, the dashboard and warehouse delivery, and the warehouse feed can land in an organization's existing analytics stack in a unified schema. The company reports 99.999% platform uptime. The privacy notice places service data storage in the United States, and the open source summary code points one model path at a US West Bedrock region.
Because the core code is public, a technical team can read how records move through the pipeline before signing. Tenant boundaries for document processing are not described, and the region and data path for every configuration, including the newer summary path and the ChatGPT plugin, are not stated.
Production access is bought through a sales conversation, with a sandbox for evaluation before any agreement. AI Summaries is a premium feature that staff turn on for an account, and the ChatGPT plugin needs MCP access enabled for the organization along with the right patient permissions. Production API use is limited to treatment requests on behalf of a covered entity.
These paths separate evaluation from the paid platform and its optional AI layer, so a team can build against the API before it talks price. No base subscription price is listed, and neither is the premium feature's rate, a charge per patient or per record query, or a separate implementation fee, so what production will cost comes only from a quote.
The platform serves outpatient and urgent care providers, health systems, record system vendors and organizations working under value based contracts. Sollis Health uses it in membership urgent care to see prior EKGs, operative reports and imaging in seconds, Color Health in cancer care at scale, Brightside Health in mental health care and knownwell in primary care and weight management, while RTHM relies on the summary for complex chronic illness.
WeightWatchers runs an obesity care configuration of the account prompt, and Amazon One Medical, Maven Clinic and Hoag round out the named customers. These are live settings with real volume. The same AI brief has not been validated across specialties, and summary error rates are not compared across settings, patient groups or record lengths.
Alternatives to Metriport
Other products that do the same kind of work as Metriport, and what each one does.
Zus Health assembles a longitudinal patient record from national data networks and delivers it to care teams, analysts and developers.
ThetaRho sells a clinical intelligence platform in three layers and will sell any of them on its own.
Abstractive Health retrieves a patient's records from the national exchange networks and summarizes them, with every sentence of the summary linked to the source note it came from.
Fourier Health automates medical record intake and turns inbound records into structured, cited clinical data and specialty summaries.
Credo Health, based in Denver and led by founder and chief executive Carm Huntress, retrieves and assembles patient records for value based care.
Predoc automates medical record retrieval and prepares the records for clinical, operational and research work. Its agents query digital sources such as health information exchanges and reach out to providers by phone or fax for records still missing.
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 |
|---|---|---|---|---|
|
Contact for pricing
|
Production access via sales. AI Summaries is a premium feature enabled on request | BAA offered. Applicable tier and AI scope are not stated | No implementation fee is listed for the Medical API or the EHR apps. | Vendor Published |
Metriport offers a sandbox for evaluation and turns on AI Summaries separately for production customers. Subscription rates, the premium feature rate and implementation fees are not listed.