It can look like one. The difference is grounding: answers are built from your approved documents and show their sources, instead of relying on what a general model happens to know.
AI knowledgeassistants.
Put what you know within reach.
We build AI assistants grounded in your own material: policies, guides, documentation, and records. Answers come with the sources behind them, access follows your existing permissions, and when the material doesn’t hold an answer the assistant says so or hands over to a person.
01 / How it worksOne possible
One possible
path.
An example workflow. Every project is shaped around your process, your tools, and the decisions your people keep.
One possible path
DocsPoliciesGuides
Refine when needed01 / The right material
Find relevant material the person asking is allowed to see.
02 / What it doesFrom the problem
From the problem
to what changes.
The problem
The answer usually exists somewhere, in a shared drive, a wiki, or an old email thread. Finding it takes time, so people end up asking the same few colleagues the same questions.
What changes
An assistant searches the approved material, answers in plain language, and links to the passages it relied on, so people can check before they act.
- 01
Grounded answers
Retrieval-augmented generation (RAG) over your approved documents, so answers come from your material rather than a model’s general knowledge.
- 02
Sources you can check
Every answer links to the passages behind it, so a person can verify it in seconds.
- 03
Access that follows permissions
People only receive answers drawn from material they’re allowed to see.
- 04
Honest gaps
When the documents don’t cover a question, the assistant says so and offers a route to a person.
- 05
Kept current
Changed documents are re-indexed on a schedule you agree, so answers don’t drift out of date.
03 / Is it a fit?The right job.
The right job.
Honestly.
A good fit
- Teams answering repeated questions from a known body of material
- Onboarding, support, policy, or product knowledge
- Situations where people need answers they can verify
Not the right fit
- Questions with no reliable source material
- Decisions that need a qualified professional’s judgement
- Material you can’t safely or legally make searchable
What you’ll bring
- The source material, and who owns it
- Who should be able to see what
- The questions people ask today
- How often the material changes
How we’ll judge it
- Answer accuracy on a test set of real questions
- Whether each citation actually supports its answer
- How unanswerable and ambiguous questions are handled
- Time to a useful answer, compared with searching today
04 / Cost & ownershipWhat shapes
What shapes
the cost.
We set out the proposed work, costs, assumptions, and milestones before you decide to proceed. Ownership, hosting, and support are agreed in the same proposal.
To build
- The volume and formats of the source material
- Permission and access requirements
- Where it lives: chat, intranet, helpdesk, or your product
- The depth of evaluation needed
To run
- Model usage per question
- Search index and hosting
- Re-indexing as documents change
QuestionsWorth asking
Worth asking
first.
05 / Let’s beginA rough idea is enough.
Your next
move. Starts here.
Tell us what could work better. We’ll help you find a useful place to start.