Agree the boundaries
Name the workflow owner, approved sources, intended outputs and human review responsibilities. Specify the number of sources, integrations and test cases in the proposal.
PRIVATE AI DOCUMENT ASSISTANTS
Find answers in your company documents, with sources your team can check.
Give your team an assistant for approved business information. Start with one document collection and a defined task, with hosting, permissions and data handling agreed before a pilot is built.
- Answers grounded in the documents your team is allowed to use.
- A focused pilot for knowledge search or documentation workflows.
- Agreed hosting, access checks and a person reviewing the result.
Tell us the document task. We reply within one working day.
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
YOUR FIRST PILOT
One useful task.
An informed rollout decision.
A bespoke pilot built around one document collection and a defined task. This is not an instant self-service subscription.
Build and evaluate
Test representative work, missing information and exceptions against agreed criteria. Compare preparation time, corrections, missed cases and the effort still needed from your team.
Hand over the evidence
Review the outputs and test results together. Document operation, known limits, review steps and the agreed support arrangements so your team knows how to use the pilot.
Decide what comes next
Use the evidence to decide whether to refine, expand or stop. Wider production rollout and additional integrations need their own agreed scope and price.
Before building: nominate a document and permission owner, identify current versions and share representative material through an agreed secure route. Hosting, supplier terms, refresh and deletion handling are part of discovery.
Tell us about your documents
Describe the task and your hosting requirements, without sharing confidential documents. We reply to discuss scope and provide a written quote before work begins.
Thank you, your enquiry has been received.
A senior consultant will reply within one working day.
- Prefer to speak now? Call 0330 043 7414.
- Add hello@nerdster.ai to your contacts so our reply is not missed.
- While you wait, read more about private ai solutions.
YOUR ENQUIRY
Start with one document workflow.
Tell us where your documents live, who uses them and which task takes too long. Our first conversation establishes the use case, data boundaries and what a useful pilot would need to prove.
WHERE YOUR AI RUNS
The right setup starts
with your requirements.
We compare these options during discovery. The proposal names the chosen environment, the services involved and the controls to be tested.
| Setup | Where processing happens | Who operates it | What to plan for |
|---|---|---|---|
| Your own infrastructure | Model processing in your environment; any external connections agreed separately. | Your IT team or an agreed support partner. | Hardware capacity, model quality, updates and backups. |
| Dedicated cloud | An agreed cloud account and region, with documented service boundaries. | Your cloud team or an agreed managed service. | Cloud costs, network controls, access and logging. |
| Managed model service | An external model provider under reviewed data-handling terms. | The provider operates the model; ownership of integrations is agreed. | Processing locations, retention, training terms and usage costs. |
BEFORE ANY WORK BEGINS
Scope and costs, agreed before you commit.
We quote for the agreed pilot. Additional integrations and data preparation are scoped separately.
The proposal sets out deliverables, responsibilities and timing. Hosting, software subscriptions, model usage and ongoing support are set out separately. Confirm discovery scope and any fee before commissioning it. VAT applies where applicable and is confirmed in your quote.
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.