Modella AI

The product is the AI. PathChat is a multimodal generative model that reasons over pathology images and text, built on pathology foundation models pretrained on histology image and image text data, and Judith is an agent for automating AI model development on biomedical images. There is no services, viewer or hardware layer: if the models are removed there is no product.

The artificial intelligence is the product.

Explicitly a co-pilot. PathChat is positioned for human in the loop clinical decision making, providing assistive discussion at region, slide, and case level plus ancillary test suggestions and slide level triage, with the pathologist retaining the diagnosis.

The generative interface raises a distinct oversight question competitors do not, since a conversational model can produce fluent but wrong reasoning, and the company frames the tool as support rather than an autonomous reader, which is the appropriate posture for a language model in diagnosis.

The approach is published in the research literature; version and update practice for the product are not. PathChat originates in peer reviewed work from the Mahmood Lab at Mass General Brigham, published in Nature (2024) and linked from the company's site, and the architecture is stated: a pathology foundation model pretrained on histology image and image text data feeding a custom trained multimodal large language model. The whole slide foundation model TITAN is described in the literature as pretrained on 335,645 whole slide images and fine tuned partly on synthetic captions generated by PathChat. Product versions are named (PathChat 2, PathChat DX), but no release history, change log or update practice is published, so a buyer cannot tell when the deployed model changed or how far it differs from the published research model.

The approach named without the version and update discipline behind it.

The model family is described, the parent is named, and no operating party is. PathChat combines Modella's pathology foundation models with a custom trained multimodal language model; which base language model it builds on is not stated on the company's pages. Modella AI was acquired by AstraZeneca, and the company says its platform will be integrated into AstraZeneca's oncology research and development organization, which makes the parent a party to any data the platform handles.

The company publishes no hosting provider, subprocessor, retention position or statement on whether customer images contribute to model development, though its privacy policy could hold some of it. Whole slide image files routinely embed a photo of the slide label, and PathChat is described as usable from a phone at the microscope, so where images go and what persists are open questions.

Published research on the model family, short of evidence from clinical use. The company links the Nature (2024) paper describing PathChat and its evaluation on pathology diagnostic questions against other multimodal models, and Nature Biotechnology work on the underlying foundation models, from the Mahmood Lab at Mass General Brigham. That is peer reviewed research on the research model; PathChat 2 is offered for research and education and PathChat DX is not cleared, so there is no body of evidence from deployment in clinical practice, and the company names no clinical customer.

Published research short of independent validation of the product sold.

Image handling is not described. The site states no retention period, deidentification practice or training use for images submitted to PathChat, though the privacy policy may address them. Three risks are particular to this product. Whole slide image files routinely embed a photo of the slide label carrying an accession number and often a name, so shared slides are identified records by default.

PathChat is described as usable from a microscope, a slide viewer or a smartphone, and phone capture means the image transits a personal device and may persist in a camera roll or backup. And after the AstraZeneca acquisition, tissue images and features derived from them are research assets to a pharmaceutical owner, and nothing published separates customer data from the parent's research use. Safety engineering for a generative model's wrong or fabricated answers is not published.

The company states no HIPAA position, and the relationship a laboratory would enter is unclear. Apart from the privacy policy, which may hold one, no page mentions HIPAA or business associate terms. PathChat 2 is offered for research and education, where institutional review board approval and research agreements, rather than business associate terms, often govern identifiable tissue.

After the AstraZeneca acquisition the counterparty is part of a pharmaceutical group, and whether a laboratory is dealing with a software vendor, a research collaborator or a data provider to a sponsor determines the agreement. A generative copilot that drafts pathology text also raises where that text lands in the record.

There is no published security attestation for the product. No SOC 2 report, ISO 27001 certificate, trust center or penetration testing statement appears on the site, whose footer links terms of service and a privacy policy. The AstraZeneca acquisition does not by itself extend the parent's security certifications to the acquired company's systems; certifications are held by legal entities, and integration often takes years.

The product is a generative system that ingests tissue images, possibly from a phone, and an agent that builds models over biomedical image data, and nothing published shows an independent examination of those systems.

A pathway stated and in progress for the clinical version, and research use for the current one. The company announced that PathChat DX, the diagnostic version of its copilot, received FDA Breakthrough Device Designation (January 2025) and states that the designation does not imply clearance or approval. The PathChat page presents PathChat 2 as a copilot for research and education use. FDA's database lists no clearance for Modella AI. Since the AstraZeneca acquisition, the company says its platform is being integrated into AstraZeneca's oncology research and development organization.

A pathway in progress, with the current product scoped to research and education.

No governance framework, evaluation method or bias analysis is published, and two structural facts matter here. The first is ownership: Modella AI has been acquired by AstraZeneca, and its platform is being integrated into AstraZeneca's oncology research and development organization.

A pathology foundation model characterizes tumor features that often decide which patients receive which therapies, and the owner develops and sells such therapies; nothing is alleged, but the company does not say how model development is separated from therapeutic programs or who decides what the models are optimized to detect.

The second is training: the whole slide model TITAN is described in the literature as fine tuned on synthetic captions generated by the company's own earlier copilot, so errors in one model can become training signal for the next with no external reference at that step. No results by patient group, site, scanner or stain are published, and hallucination risk in the generative copilot is not addressed.

The pathologist decides, and nothing stands behind the copilot's answers. PathChat is presented as a copilot for research and education, with the pathologist or researcher judging what it says; the clinical version, PathChat DX, holds a Breakthrough Device Designation but is not cleared. The Nature paper the company links evaluated the research model on diagnostic questions, which describes a research system rather than the product a laboratory would use, and no error or hallucination rate is published for PathChat 2 or PathChat DX. A generative assistant produces prose a pathologist may carry into a report, and a fabricated or subtly wrong statement about morphology reads like a correct one. No warranty, indemnity or remediation commitment is published.

Human judgment over every answer, with nothing attached to the output.

No production integration with a laboratory system is named. PathChat is presented as usable from a microscope, a slide viewer or a smartphone, which lets a pathologist consult it without a completed digital pathology program. The same design leaves it outside the laboratory information system: nothing it produces is captured against a case, no audit trail links a consultation to a report, and generated text reaches the record only if someone copies it. Research collaborations announced with Techcyte, whose platform hosts partner algorithms, and with illumiSonics point toward platform integration, but they were described as research rather than shipped products.

Integration claimed through collaborations with no system named.

Nothing published says where inference runs. The company's pages name no hosting provider, region, tenancy or residency term, and only the privacy policy might. A generative vision language system typically needs accelerated infrastructure few laboratories operate, and the phone and microscope capture routes the company describes imply that images are sent somewhere for processing.

The Judith agent automates model development on biomedical image data, which implies an environment where data is used to build models, and after the AstraZeneca acquisition whether infrastructure stays with the subsidiary or moves into the parent's systems is not stated.

No product pricing is published, consistent with a research use tool and a pre market clinical version. The company is venture backed and private, so there is no listing driven financial disclosure either. Commercial terms are opaque at this stage.

Concentrated on anatomic pathology, where the generative co-pilot approach is applied across region, slide, and case level reasoning with ancillary test suggestions, making it broad within pathology rather than tied to a single tumor type. That generality is the point of a foundation model, distinguishing it from single indication pathology detectors, but coverage is confined to pathology and the clinical version is not yet available.