Retrieval over your own documents
Answers grounded in the knowledge base you connect, with confidence and coverage metrics and provenance on every answer. The sources and the index stay on the appliance.
Semantic retrievalNeuralSeek and IBM Fusion
behind your firewall
NeuralSeek runs on IBM Fusion HCI as a single appliance in your data center, with retrieval, agents, guardrails and governance included. Your data stays on site.
IBM FusionWhere NeuralSeek fits
How NeuralSeek runs on IBM Fusion
Your data center holds the rack. IBM Fusion provides the hardware and the container platform, NeuralSeek runs on that platform, and your teams use what is built on top. Every layer sits in the same room, so your data stays there too.
Your assistants and agents
Search, answers, drafting and multi-step agents, the tools your teams open every day.
NeuralSeek
Retrieval over your documents, agents, guardrails and governance. The same container that runs in the cloud, here on your hardware.
Red Hat OpenShift
The container platform NeuralSeek installs onto. IBM describes Fusion as pre-configured infrastructure for OpenShift, so it is ready on arrival.
IBM Fusion HCI
Compute, storage and networking in one system, with GPU-powered nodes for inference, in your data center beside your identity provider and your data.
How it is deployed
From rack to answers
Your documents, prompts and logs stay inside your building at every step.
The Fusion rack goes into your data center
IBM Fusion HCI arrives as pre-configured infrastructure for Red Hat OpenShift: compute, storage, networking and GPU nodes in one system. It joins your network and identity provider like any other system.
NeuralSeek is installed on OpenShift
NeuralSeek on-prem runs as containers on OpenShift or Kubernetes, so it goes onto the cluster Fusion already provides, with no extra servers to buy.
Connect your documents and a model
Point it at the document stores and systems your teams already use, and at the language model you choose. The index lives on the appliance, next to the sources.
Build, govern and run
Teams build assistants and agents in the console. Guardrails check every answer and every run is logged on the appliance, which can run fully air-gapped.
We scope sizing and installation for your environment, starting from what you already run.
Governed on arrival
Guardrails come with the appliance
The rules for what an answer may say, and what it must refuse, run in the same unit in your data center. You configure them in the product and they apply on every run.
console › guardrails
What NeuralSeek adds
What NeuralSeek adds to Fusion
Fusion supplies the infrastructure. NeuralSeek adds retrieval, agents, guardrails and governance.
Every capability above ships in the same container wherever it runs. The appliance gives it a home in your data center.
Agents built on the platform
Multi-step workflows that query documents, databases and APIs, written in NeuralSeek's template language without traditional coding.
Agent workflowsGuardrails on every request
Screening, redaction and grounding checks run inside the appliance before and after every model call.
GuardrailsGovernance you can show an examiner
Dashboards for confidence, coverage, usage and cost, and a log of what was asked, retrieved and answered, kept on the appliance and exportable to the collector you already run.
Governance
Where it is used
For work that stays on site
One appliance for regulated industries. Records stay on the appliance and off public networks.
Healthcare
Patient data stays in the hospital
For hospitals, payers and the teams that write clinical and medical-education content.
- Back-office assistants with patient identifiers masked before any model call
- Clinical research answers, every citation traceable to its document
- Patient-services chat with every interaction logged for review
- Medical education and regulatory drafting from grounded sources
Financial services
AI your examiners can inspect
For banks, payments, pensions and investor research, with the model, the data and the audit trail on your own hardware.
- Customer-facing banking assistants, with replay on every response
- Internal search over policies and procedures
- Financial-crime and onboarding agents behind your own guardrails
- Loan files, mortgage packs and filings processed on site
Energy and utilities
AI for grid and field teams
For grid operators, field crews and the regulatory teams that support them.
- A grid-operations copilot over switching procedures and runbooks
- A field-crew agent grounded in manuals and safety bulletins
- Outage triage and customer updates from the same operational record
- Regulatory drafting and searchable permits and asset records
Next step
Learn more
The appliance is one way to run NeuralSeek. More on how it works.
The IBM solution paper
IBM's paper on NeuralSeek and IBM Fusion HCI: the problem regulated organizations face, what each side provides, the architecture and sizing guidance.
Download the PDF from ibm.comNeuralSeek on IBM Cloud
Our SaaS through IBM: the catalog entry, watsonx as the model and knowledge layer, and security review answers.
See the IBM Cloud pageNeuralSeek and watsonx
How NeuralSeek works alongside IBM's own AI platform, and why IBM customers often run both.
Read the comparisonDeployment
What is in the container, how to buy it and what running it asks of your team.
Deployment optionsSovereignty
How NeuralSeek keeps data and AI inside your jurisdiction.
Data and AI sovereigntySecured AI: the webinar
The September 2026 session on IBM Fusion and NeuralSeek, with a summary and the recording.
Watch the session
Tell us about your environment and we will scope the appliance with you.
Start building with NeuralSeek today.
Sign up online and start with a two-week trial, or talk to a NeuralSeek expert about running it in your own cloud.