Your EHR holds years of clinical history. The question is whether you can actually access it.

A patient arrives for a follow-up appointment. You have 20 minutes. Their file spans four years, a hospitalization, a crisis episode, three medication changes, and a referral letter  from 2022 (in a pdf without title ofcourse). You know the information is in there. You just don't have time to find it. This is not a documentation problem. It's an access problem. And it's one that AI can actually solve: if AI is built to work safely inside your EHR, rather than somewhere outside it.

The EHR as a locked library

Over years, EHRs accumulate enormous clinical value. The data is there. The problem is that retrieving it (quickly, reliably, during a consultation) can be surprisingly hard. One reason for this is that most EHR systems are designed for documentation, not for retrieval. You can navigate to a specific note if you know where to look. What you can't do is ask a question and get a synthesized answer drawn from everything in that patient's file. 

Search and summarize: not just another AI feature

Delphyr is designed to close this gap. Our clinical AI platform integrates directly into your EHR. One of its  core capabilities — Search and Summarize — lets clinicians ask plain-language questions (in a chat environment added to your EHR) and receive answers sourced from the patient's actual record.

Ask: "What happened during previous crisis episodes?" and you get key events, risks, and outcomes — pulled from notes and reports, condensed, cited. Or, ask: "How has this patient's medication changed over the last five years?" and you get a timeline, not a list of file names to open one by one. Or want to know what's essential to know before today's intake? Just ask it and you get the relevant history, without scrolling through hundreds of pages.

Why this is different from using an open AI model 

When people first hear about Delphyr, a reasonable question is: why not just use ChatGPT? It can summarize, too. The answer comes down to architecture (and what that architecture means for safety). General-purpose AI tools like ChatGPT, Claude, and Gemini generate answers from patterns learned during training, not from your patient's record. When you paste a note into ChatGPT and ask for a summary, the model works from what it's seen before, with no connection to your organization's systems, your clinical guidelines, or the rest of that patient's history.

That creates two problems. First, anything you paste in has just left your organization's secure environment. Patient data has been transferred to a third-party commercial platform not certified for healthcare use. Second, the model can only work with what you give it. It has no access to the other four years of records sitting in your EHR.

Delphyr works entirely differently. It never moves data outside your EHR. Instead, it searches the record in place, retrieves the most relevant content, and builds its response from what's actually there. This approach (called Retrieval Augmented Generation, or RAG) means Delphyr's answers are grounded in your patient's specific history, supplemented by clinically vetted sources, not information from the open internet.

What makes it reliable

To make sure Delphyr gives clinicians reliable answers, it is designed with many safeguards that general AI tools don't offer: For example, every response includes citations and references back to the specific documents in the patient's record that the answer draws from. This means a clinician can verify the output in seconds rather than taking it on trust.

Also, the system is designed to say "I don't know" rather than generate a plausible-sounding answer when the record doesn't support one. That's a deliberate design choice. In healthcare, a confident wrong answer is worse than no answer.

These are just a couple of examples in how we make our AI safer, but the results bear out. On HealthBench Professional, one of the most rigorous clinical AI benchmarks available, Delphyr scored 58.3%, placing it second overall (first place was GPT 5.4 in ChatGPT for Clinician with a score of 59%) , ahead of every general-purpose frontier model. In clinical writing and documentation specifically, Delphyr scored 65.4%, substantially higher than GPT-5 variants. These are precisely the tasks that search and summarize supports: synthesizing patient notes, surfacing relevant history, producing structured clinical output. (Full benchmark results here.)

Safety is built in, not bolted on

Because Delphyr operates with patient data, compliance isn't optional, and it isn't an afterthought. Delphyr is designed for full GDPR-compliancy, ISO 27001, and NEN 7510, the Dutch information security standard for healthcare. It is also developed in line with the EU Medical Device Regulation (MDR), which governs safety and quality standards for software functioning as a medical device. Patient data is processed within the European Union and is never used to train AI models. Delphyr does not retain data after a session ends.

The role of the clinician doesn't change

Quickly searching and summarizing within a patient file doesn’t replace clinical judgment. Delphyr surfaces information, and what you do with it remains your professional responsibility. That means checking the output. Verifying that what the system surfaced is accurate, complete, and relevant to the specific patient in front of you. AI-generated summaries are a starting point, not a sign-off.

Moreover, it means being deliberate about where AI tools fit in your workflow. Delphyr belongs inside your EHR, where it has context and access. General-purpose AI tools belong outside clinical use, and patient data should never be entered into them.