Thalix Health

What Thalix is

What is Thalix?

Thalix is a hospital integration and AI agent platform built by Zapx Labs, and Thalix Health is the consultancy that uses it to hand hospital work to AI agents safely. The platform connects hospital systems and devices into one ordered, replayable stream of events, serves that data to people, apps and agents under one policy, and audits every read and action. Agents draft, check and route work; people make the decisions.

What does Thalix replace, and what does it not replace?

It can take the place of the plumbing under every dashboard and assistant: the integration layer, the patient index linking identifiers across systems, and the audit log. It does not replace the EPR, PAS, PACS or lab system. It connects them and serves their data under one set of rules.

Which parts of the hospital is Thalix for?

Any work that is repetitive, follows rules or is spread across systems, where an agent can prepare it and a person decides. The home page lists clinical care (results follow-up, letters, handover), patient flow (beds, discharge, transfers), diagnostics, pharmacy, outpatients and referrals, theatres, administration and finance (coding, claims, rotas), and quality and research. The demo hospital runs bed turnaround, abnormal result follow-up and discharge.

Is Thalix a medical device?

No. The Thalix demo is not a medical device, and the platform is designed so that agents never make clinical decisions. Clinical decisions and orders (tier 3) exist only as signed, locked apps from a trusted publisher and always need a second person. If a hospital wanted Thalix to carry software with a medical purpose, its regulatory status would have to be settled first.

Integration: HL7 v2, FHIR, imaging, devices and files

What is a healthcare integration engine and how does Thalix's work?

An integration engine sits between hospital systems: it receives their messages, checks and acknowledges them, transforms and routes them, and holds what it cannot process. In Thalix, one driver per protocol turns each incoming message into a standard event with FHIR R4 content, links the patient identity, drops duplicates, keeps the raw message and writes it to an ordered, replayable log. A sender is acknowledged only after its event is durably logged.

Which HL7 v2 messages does Thalix support?

HL7 v2.3 to v2.8 over MLLP, with TLS or mutual TLS: ADT, ORU results (lab, radiology, pathology and monitor vitals), orders (ORM, OMG, OML), scheduling (SIU) and documents (MDM), mapped to FHIR R4. It can also send HL7 v2 to a receiver. Batch files, enhanced-mode acknowledgements, pharmacy (RDE, RAS), immunisation (VXU) and billing (DFT, BAR) messages are not mapped yet; they are kept as unmapped events with the raw message. A message that cannot be mapped is never dropped: it is kept as a dead letter with the raw message, ready to replay after a fix.

Does Thalix support FHIR?

Yes. Clinical data inside Thalix is held as FHIR R4. The fhir driver talks to FHIR R4 servers with read, search, create, update and transactions, polls for changes, receives R4B and R5 topic subscriptions, and authenticates with SMART Backend Services.

Can Thalix connect to PACS, bedside monitors and other devices?

Yes. A DICOMweb driver reads new and updated studies from a PACS or VNA as FHIR ImagingStudy records. An MQTT driver takes vitals, location tracking, bed and nurse-call sensors through declarative mapping rules, and IEEE 11073 SDC devices connect through a gateway on the external driver protocol. Webhooks, CSV and NDJSON drop folders, FHIR bulk files and drivers written in other languages are also supported; DICOM C-STORE is on the roadmap, not built.

Can Thalix work alongside our existing integration engine?

Yes. The HL7 v2 driver is built to receive the feeds an existing integration engine already sends, so a pilot does not have to rewire the hospital. Thalix can also deliver events onward to any HTTP endpoint, at least once.

How does Thalix match the same patient across systems?

Every identifier a source sends, such as a hospital number, NHS number, FHIR id or bedside device assignment, links to one Thalix patient. Identifiers that already belong to different patients raise a conflict for a person to review instead of being merged automatically. Merges are explicit, audited calls that record who merged what and why.

What is the difference between an integration engine and an API gateway?

An API gateway sits in front of APIs and applies the same controls to every call, such as authentication, rate limiting and logging. An integration engine moves and reshapes data between systems, including ones that cannot call an API. Thalix is on the integration side; it serves its own API and an MCP endpoint, but it is not a general-purpose gateway for a hospital's other APIs.

Agents, workflows and people in control

What is an AI agent in Thalix?

An assistant that works through the Thalix kernel with the same tools, policy, redaction and audit as any other user. In a conversation it acts for the signed-in person and never sees or does more than they can. It can search the record, read hospital documents with citations, keep scoped memory and use approved external tools, and it proposes actions for the person to confirm.

Can an agent act on a patient?

Only as a proposal, for the person it works for. The person confirms the card, the action runs as them, and the kernel's approvals still apply. Agents can never run or approve tier-3 actions such as clinical orders, and a tier-3 order always needs a second person.

Which AI models does Thalix use, and where do they run?

On site by default, through Ollama with open models on the node or a hospital server. Hosted models can be added: Anthropic's Claude and OpenAI-compatible endpoints such as OpenAI, Azure OpenAI and vLLM. Every model declares the data classes it may receive (operational, demographics, clinical; operational only by default), and the kernel enforces that ceiling on every call, so a cloud model sees operational data only, never patient records.

What are Thalix workflows?

A workflow is a hospital process written down so the platform can run it: steps that create tasks for roles or people with forms, due times and escalation, ask agents for drafts, wait for something to happen, and branch on what happened. Workflows run only low-risk operational actions such as bed states and tasks by themselves; every other action is confirmed by a person, who runs it as themselves. Workflows never see patient names.

How does Thalix help with bed management and discharge?

In the demo, HL7 v2 ADT from a simulated hospital keeps Ward 4's board live. The Discharge workflow runs medicines to take out, transport and a drafted summary in parallel, a doctor signs the summary, and when the patient leaves the bed goes to cleaning. Bed turnaround then asks housekeeping to confirm the bed is clean, frees it, and escalates to the nurses if it takes more than 20 minutes.

How are new agents, workflows and boards approved?

They ship as versioned apps with a risk tier worked out from their contents: 0 views, 1 operational, 2 clinical information, 3 clinical decisions, with the reason shown. Nobody approves their own app, and tier 2 also needs the clinical safety officer's recorded sign-off. Tier-3 apps exist only as signed, locked packages from a trusted publisher, and a failed upgrade puts the previous version back.

What stops an agent being tricked by text in a document or message?

The protection is structural, not just the prompt. Tool results, documents and messages are treated as data, and whatever a model is persuaded to do, it can only call its listed tools under the user's rights and its data ceiling. Actions become proposals for a person to confirm, and everything is audited.

Data, privacy, security and audit

Where is patient data stored?

On the hospital's own servers. Every Thalix store is a SQLite database inside the node's data volume, so there is no separate database to install. The data directory holds patient data and should sit on an encrypted volume restricted to the service account, with retention set to the hospital's records policy.

Who can see what in Thalix?

Every request is checked against the hospital's policy: who is asking, for what purpose, their unit and care-team relationships, and which fields they may see. Fields are redacted by class (operational, demographics, clinical), so a porter and a nurse see the same ward differently. Break-glass access for clinicians is time-limited, needs a reason, is audited and creates a review task for the privacy officer.

What does Thalix audit?

Every read, search, stream, action and denial, by people, agents and workflows, in a SHA-256 hash-chained log that can be verified from the workspace. Agent runs are recorded as the agent and the person it acted for. For any entry, the kernel can name the policy that decided it.

How is Thalix secured?

Network listeners stay on loopback until TLS 1.2 or 1.3 is configured, and can require mutual TLS. API access uses bearer tokens with roles, compared in constant time, and secrets are held as references to environment variables or files, never in the config. Patient data is kept out of logs, errors and metrics, and raw messages are shown only to administrators who state a reason, with every reveal audited.

Does Thalix work with our single sign-on?

Yes. Thalix accepts OIDC access tokens from the hospital's identity provider, such as Entra ID, Okta, Keycloak or Ping, as well as its own kernel tokens for services. Agents act for a signed-in user and never see more than that user.

Deployment and hardware

How is Thalix deployed?

From one Docker Compose stack: the Thalix node, Ollama with pinned local models, and optional MQTT and NATS brokers, with HTTPS certificates made on first start or taken from the hospital's own certificate authority. Production configuration covers the identity provider, drivers, models and apps, with secrets passed as environment variables.

Can Thalix run without internet access?

Yes. With local models it keeps working offline, and for sites without internet a single offline bundle carries the images, model weights, Docker Engine for Linux, an installer and checksums. The installer checks the checksums, loads everything and starts the stack.

Can Thalix come on its own hardware?

Thalix in a box is a concept, not a shipping product. The home page shows Edge, Site and Rack designs as renders made with an image model, not photographs, and their specifications are illustrative. If you would rather not provision servers, tell us about your site and we will size it properly.

Clinical safety and assurance

Is Thalix deployed in any NHS trust?

No. Thalix is at pilot stage, and every demo runs on synthetic patients. A hospital deployment needs its own clinical safety case, information governance approval and a security review first.

Does Thalix have DTAC, DSPT or DCB0129 sign-off?

No. Thalix has no completed DTAC assessment, no DSPT status, no DCB0129 clinical safety case and no other certification, and we do not claim otherwise. If a pilot gave us access to NHS patient data or systems, completing the DSPT and the DTAC form for that version would be part of the work.

What would a hospital deployment need?

A clinical safety case (in England, DCB0129 from the manufacturer and DCB0160 from the hospital), information governance approval, and penetration testing as part of a security review. Thalix's own security notes list these, and development and demos use synthetic data only.

How is our clinical safety officer involved?

From day one of a pilot. In the platform, the clinical safety officer's review is a role, not an email: tier-2 apps, such as agents that read patient data or actions that write to other systems, cannot be installed without that recorded and audited sign-off. Your clinical safety team reviews each change before it goes live.

What evidence can Thalix give a safety case or security review?

Concrete records rather than promises: the hash-chained audit of every access, the policy that decided each one, agent runs recorded as the agent and the person, risk tiers with their reasons, signed and locked tier-3 packages, rollback on failed upgrades, and dead letters that show what could not be processed. The hazard log and safety case themselves remain your clinical safety officer's documents.

Pilot and getting started

How does working with Thalix Health start?

With a 45-minute conversation at no cost about what is hurting, what you have tried and what good would look like. Then an assessment of a few weeks with one department or pathway, a pilot of one workflow with your clinical safety officer involved from day one, and finally extending to more departments and training your people to build, review and run their own agents and workflows.

How much does Thalix cost?

There is no public price. Talk to us about a pilot and we will scope it around the work, the systems involved and your site.

Can I see Thalix working?

Yes. The home page has a narrated three-minute walkthrough of a fictional demo hospital with synthetic patients, and we can run the synthetic hospital with you: feeds arriving in one log, the same patients seen differently by role, an assistant that proposes while a person confirms, a result follow-up workflow, the ward board and the audit chain. Get in touch through the contact form on thalixhealth.com or at hello@thalixhealth.com; we reply within two working days.

Watch it, and longer answers

Watch the 3:08 Thalix walkthrough, with a full transcript.

The guides explain the standards behind these answers (HL7 v2, FHIR, DCB0129/DCB0160, DTAC, the DSPT) with official sources linked.

Thalix Health

See Thalix on a demo hospital.

A conversation with us is 45 minutes and free: the work you want to hand to agents, the systems involved, and where Thalix would and wouldn't fit. Demos run on synthetic patients.