Find the right information
Ask a question across approved policies and procedures. See the source behind the answer, with document permissions built into the design.
YOUR INFORMATION. YOUR BOUNDARIES.
Your company knowledge. Your own AI assistant.
Help your team find answers, prepare drafts and manage compliance documentation using approved company information. We build around your hosting requirements, document permissions and human review process.
Let’s discuss Private AI solutionsDOCUMENT WORKFLOWS
Documentation with clear review responsibilities.
A document assistant can help organise evidence and prepare material for review. Its role, approved sources and limits need to be defined and tested for your organisation. It does not confer regulator approval, a CQC rating or UKAS accreditation.
Whether you need to find information, prepare a review pack or draft a response, we agree what the system should do and where a person needs to check the work.
Discuss your documentation workflow
WHAT IT COULD LOOK LIKE
Ask a question.
See the source.
Keep the final say.
A useful assistant should make its evidence easy to inspect and tell you when information is missing.
Static mockup with fictional documents, not a live product demo or customer result. The source panels below show sample text; they do not demonstrate a working retrieval or access-control system.
Document assistantMockup
What is missing from our supplier review pack?
ANSWER FROM THE SELECTED DOCUMENTS
The checklist asks for a supplier questionnaire and a review record. I found the questionnaire, but no completed review record in the selected documents.
Needs a person to check
A missing search result does not prove a document is missing. Ask the pack owner to confirm.
Inspect the example sources
Supplier review checklist · v2.1
Fictional excerpt: “The review pack contains the supplier questionnaire and a completed review record.”
Supplier questionnaire · v1.0
Fictional document: a completed supplier questionnaire. A questionnaire alone does not establish that the review is complete.
When the answer is unavailable
Question: Who signed off the review?
Example response: I cannot establish that from the selected documents. Ask the review owner; do not infer approval.
When a role cannot access information
Intended response: I cannot provide information outside your approved sources. Contact the document owner about access.
This is illustrative wording. Your pilot must test that restricted content is withheld across answers, references and logs.
Draft answer · review before use
Manage compliance documentation
Scope a workflow for organising evidence, identifying missing documents or preparing a review pack for a named owner.
Prepare a first draft
Use approved material to prepare a document or response, with references and a review step before anyone relies on it.
THE RIGHT STARTING POINT
Start with what needs to stay private.
Private AI starts with clear boundaries. We map the information involved, compare deployment options and agree how prompts, documents, outputs and logs should be handled. Then we test one useful task with your team before considering a wider rollout.
The outcome
A scoped private AI pilot and a documented view of its data flows, controls, costs and limitations, so you can make an informed rollout decision.
WHAT WE LOOK AT
The detail that
makes the difference.
We connect the technical decisions to the people, information and processes involved in your everyday work.
Where it runs
Compare infrastructure in your own environment, a dedicated cloud deployment and a managed model service. Document which suppliers process information and which connections are required.
What it can access
Choose approved sources and user roles. Test that the assistant respects document permissions, including when users ask questions across multiple sources.
What happens to your data
Agree handling for prompts, uploaded files, retrieved passages, outputs, logs and backups. Record retention, deletion and model-training terms for the chosen setup.
How people use it
Start with internal knowledge search, document drafting or one repeatable workflow. Test answer quality, source references, access boundaries and escalation to a person.
HOW WE WORK TOGETHER
A considered process.
A clear next step.
Define the boundaries
Agree one use case, its owner, approved information and the hosting and access requirements.
Build and test a pilot
Configure the agreed environment and test useful answers, permissions and failure cases with representative examples.
Hand over with a plan
Review the evidence, train the team and document ownership, maintenance and the decision on wider rollout.
A scoped private AI pilot and a documented view of its data flows, controls, costs and limitations, so you can make an informed rollout decision.
GOOD QUESTIONS
A little more
clarity.
What do you mean by private AI?
An AI setup with explicit controls over hosting, access and data handling. The scope can range from a dedicated cloud environment to a model running on your own infrastructure. We document the boundaries and any external services involved before building.
Will our information leave our environment or train a model?
That depends on the selected architecture and supplier terms. We agree where processing happens and check handling of prompts, documents, outputs and logs, including retention and training use. If information must stay within your environment, that becomes a requirement for the design and testing.
Can we start with our own documents?
Yes. A first pilot can help staff find answers in an approved document collection, with source references and access checks. We agree which documents are suitable and test what happens when an answer is missing, ambiguous or restricted.
Can it run on premises or without internet access?
We can assess those requirements during discovery. Hardware capacity, model suitability, integrations, updates and ongoing operation all affect feasibility and cost. An offline deployment needs its own design and acceptance tests; it is not implied by the word private.
What will it cost and who maintains it?
We quote after agreeing the use case and deployment requirements. The proposal separates implementation, infrastructure or model usage, and ongoing maintenance. We name the people responsible for updates, access reviews, backups and support before handover.
What documents and people do we need for a pilot?
Choose one current document collection, with a business owner who can check answers and a permission owner who can approve access. Identify duplicates, obsolete versions and material that must be excluded. The source owner remains responsible for keeping business information current.
What happens when documents are updated or deleted?
We agree how source changes reach the assistant and test refresh and deletion behaviour. The design must address search indexes, stored copies, references, logs and backups; a deletion in the original folder does not by itself prove every copy has gone. Refresh timing and retention depend on the chosen setup.
Who owns the setup and what support is included?
The written agreement needs to specify code and configuration rights, account ownership, operating responsibilities, handover and support coverage. Supplier licences and hosting arrangements affect these terms. Ongoing support is priced separately from the pilot.
Is this a subscription we can activate today?
No. This is a bespoke pilot: we first agree a document task, hosting, permissions, review responsibilities and evaluation criteria. Discovery scope and any fee need to be agreed before that work begins.
What if the pilot does not meet its criteria?
Review the evidence before a rollout decision. We can discuss refinement, a different approach or stopping; the project agreement must define included rework, any further fees and cancellation terms.
LET’S FIND YOUR NEXT STEP
Tell us what you
want to improve.
We’ll listen, ask a few useful questions and help you work out where to begin.
Or go straight to the private ai solutions enquiry page