Company knowledge assistant
One assistant that answers employees’ questions from your files, transcripts and databases, shows where each answer came from, and only uses what that person is allowed to see.
Your company already has the answers. People just cannot find them, and the wrong people can sometimes see the wrong things.
Searching eats the week
Knowledge workers spend nearly 20% of their workweek looking for internal information or tracking down a colleague who knows. It is everywhere: policies in PDFs, how-tos in videos, numbers in databases and spreadsheets, decisions in chat threads.
McKinseyThat time is expensive
US private employers pay an average of $46.89 per hour worked in wages and benefits.
Bureau of Labor Statistics, June 2026Generic AI tools ignore permissions
Paste everything into one chatbot and an intern can ask about salaries or a client’s contract. Most teams do not try at all for that reason.
What the agent does
- 01
Brings your information together
Documents and policies go into knowledge bases. Videos and meetings are added as transcripts. Live numbers come from your databases through read-only tools, so it can answer “how many open orders in Texas?” from the real data.
- 02
Separates it by team
HR, Finance, Sales, Engineering and each client get their own knowledge base or collection.
- 03
Controls access per question
Your app tells Studio who is asking. Each request carries that person’s ID and the collections they are cleared for, so the agent only searches what they are allowed to see.
- 04
Shows its sources
Every answer lists the documents it opened, so people can check it and trust it.
- 05
Routes to the right expert
A front assistant can pass questions to department agents (HR, IT, Finance) that it calls as tools.
Checked before anyone sees an answer
The guardrail
Access is decided per question. The agent can only search the collections the person asking is cleared for, so a sales rep’s question never touches HR files. A separate guardrail blocks requests for things nobody should get, such as passwords and personal data.
Where a person steps in
There is no approval queue in this use case. Every answer lists the documents it opened so the person asking can check it, and your own app decides what each person is cleared to see.
What is recorded
Every question, answer, source and cost is logged, and can be filtered by user, team or app.
The steps, in order
Start with one team’s documents and have it working the same day; add teams, databases and permissions one at a time.
| Step | What you do in Studio | Typical time |
|---|---|---|
| 1Pick a first team | Choose the team with the most repeat questions, often HR or IT. | A decision |
| 2Add their content | Create a knowledge base; add policies, how-tos and video transcripts as folders. | 1–2 hours |
| 3Create the assistant | Give it a name, a goal and a model, attach the knowledge, and set an output schema with sources. | 15 min |
| 4Test it | Ask 20 real questions in the Playground and save them as a test suite. | 30–45 min |
| 5Put it where people work | Use the API from your intranet or chat tool, or a web embed. | 1 hour or more, depending on the tool |
| 6Add access control | Add more teams as separate collections; your app sends each person’s cleared collections with every question. | About a day for a developer |
| 7Connect databases | Add read-only tools for the systems people ask about most. | Per system |
A worked example
A 100-person company. The inputs are assumptions about a typical business; replace them with your own numbers.
| Line | Assumption | Per year |
|---|---|---|
| Time spent searching today | About 20% of a 40-hour week, which is 8 hours per person | 800 hours a week across the company |
| Time the assistant saves | 1 hour per person per week, an eighth of that search time, over 48 weeks | ≈ 4,800 hours |
| Value of that time | 4,800 hours at $46.89 an hour | ≈ $225,000 |
| Fewer interruptions for experts | HR, IT and senior staff answer fewer repeat questions | Not counted |
| Total value | ≈ $225,000 | |
| Krazimo Studio cost | ≈ $6,500–$8,000 a year: the Growth plan at $499 a month (40,000 runs) plus model usage. A team of 20 starting on one department fits the Starter plan at $99 a month | |
About an hour per person a week, which is 4,800 hours a year across the company, and new hires get up to speed without waiting for someone to answer.
See it working
on your own information.
Tell us about your business and we will walk through this use case with you in Studio.