AI Agents · Internal knowledge assistants
An assistant that answers from your own documents.
Policies, procedures, price lists and handover notes exist, but nobody can find them. A knowledge assistant searches the documents each person is allowed to read, answers from them, and shows where the answer came from.
The problem
Where the time goes today.
- Staff ask colleagues because searching the shared drive does not work.
- Different people give different answers to policy questions.
- New joiners take months to learn where things are written down.
- Confidential documents that must not leave the premises.
How it works
What the agent does, step by step.
- Index approved documentsOnly the folders and versions you choose, kept up to date.
- Check who is askingRetrieval is limited to documents that person may read.
- Retrieve the relevant passagesThe agent finds the passages that bear on the question.
- Answer only from those passagesAnd cite them. If nothing relevant is found, it says so rather than guessing.
- Log and improveUnanswered questions show what documentation is missing.
Integrations
What it connects to.
“Used in our projects” means we have built and run that kind of connection. Anything else needs a feasibility review before we commit.
| System | Status | Note |
|---|---|---|
| Document stores and file shares | Proposed: feasibility review | Depends on where your documents live. |
| Google Workspace and Microsoft 365 | Proposed: feasibility review | Needs admin consent and a feasibility check. |
| Offline or on-premise models | Used in our projects | Odontofusion OS runs a dental-trained AI offline. |
| Existing intranet or wiki | Proposed: feasibility review | Via export or API. |
In practice
Examples of the work.
Illustrative exchanges with fictional data. They show intended behaviour, not recorded client conversations.
Policy question
Staff: “What is our policy for a remake?”
Assistant: states the policy, citing the document and section it came from.
Summarise a document
Manager uploads a supplier contract. Assistant summarises the key dates and obligations and flags where it is unsure. A person still reviews the contract.
Security and control
What keeps it safe.
- Permission-aware retrieval: no document is used for someone who could not open it.
- Answers are grounded in the retrieved text and cite sources.
- Option to run fully on-premise or offline.
- Documents are not used to train public models.
- Retention and deletion follow your policy.
Implementation
How we would approach it.
- Choose the corpusWhich documents, who owns them, how they are kept current.
- Index and evaluateTest with the questions staff really ask.
- Tune permissionsVerify that restricted documents stay restricted.
- Launch and maintainTrack unanswered questions and keep documents fresh.
Commercial considerations
What it costs, and what it saves.
- Scoped by document volume and where it must run.
- On-premise deployment needs hardware; cloud deployment has usage costs. We price both.
- Quality depends on the documents. Gaps in your documentation will show up as gaps in answers.
Build this agent with Odontofusion.
Tell us the job. We will tell you honestly whether an agent is the right tool for it.