KuikAgents builds the system connections, memory, permissions and approval flow around the model. You choose the model; your agents work safely in your systems.
Connect your procedures, contracts and regulations to the Document Vault; scanned PDFs are read too. Before answering a question, the agent finds the relevant documents, bases its answer only on them, and shows which clause of which document it came from. Everyone gets answers only from the documents they are authorized to access.
Every team builds up unwritten knowledge over the years. KuikAgents captures it before it is lost: preferences, exceptions, lessons from past work. A new employee and a newly set-up agent both start with this knowledge. You see and edit what is in memory, and you set the sharing rules. Available in the Advanced and Enterprise packages.
The mirror is a regularly refreshed copy of the tables you choose. Questions that scan thousands of records, like “What is our revenue this year?”, run on this copy instead of the live system and get answered in seconds. Checkout, warehouse and store operations continue without interruption all day; write operations go to the live system directly, with approval. Available in the Advanced and Enterprise packages.
You work with your own model key; model costs are directly between you and the provider. Simple tasks run on a lightweight model, complex tasks on a powerful one.
Your team does not need to learn a new app. Store staff reach the same agent on WhatsApp, head office on Slack or Teams, and managers on the web. Task questions can be answered from email with one click; the agent picks up where it left off.
Daily stock watch complete: 187 items scanned, 34 need attention.
@Store Manager send the list for Moda
9 items are at a critical level in Moda. List attached, transfer suggestions marked.
In the mobile app you chat with your agents and follow your dashboards and tasks. When an agent prepares an action that needs approval, you get a notification; you approve or reject it right from the notification, without opening the app. You can also ask your question by voice instead of typing it.
With the desktop app, the agent can read Excel, Word, PowerPoint, PDF and CSV files, only in the folders you allow. It uses a price list or supplier quote that was never uploaded to the system directly in the chat. It also works on the websites you allow in your browser; form submissions always wait for your approval.
A language model can calculate the same debt correctly three times and differently the fourth. In KuikAgents, calculations such as late-payment charges, account balances and bonuses are turned into a function once and run with the same code every time. The model only interprets the result; since it does not rethink the calculation each time, token costs drop too. Available in the Advanced and Enterprise packages.
Agent’s suggestion: This task includes a late-payment charge calculation. Adding a deterministic function here would make the result the same on every run and lower token costs.
When an agent completes a task correctly for the first time, the steps it followed are pinned. Every later run follows the same steps; results stay consistent and token costs drop. You see all runs on the calendar.
In group chat, several agents work in the same conversation. The system picks who answers, or you name the agent with @. A question that spans several areas is split between agents.
@Finance Manager and @Purchasing Manager, is there enough cash for the winter season purchase?
In the 8-week projection, if the purchase is made in one go, cumulative cash turns negative in week 3.
Splitting the purchase into two waves based on supplier lead times closes the cash gap. I have worked out the amounts per wave.
Score test scenarios with exact match and a judge model. Every task run is checked against its own goal; when there is a deviation, you see it.
Start right away with sample data, or let’s set up an agent together in a demo with one of your own systems.