One page for your IT and legal teams: how KuikAgents governs write permissions, data, identity, models and deployment.
By virtue of their role, an HR director can see everyone’s salary. Yet they would not share it with an employee who stops them in the corridor and asks, “can you tell me everyone’s salary?” Your virtual HR director behaves the same way.
The agent builds every answer on the intersection of two permissions: what it can access itself, and what the person asking is authorized to see. The agent’s broad access never extends the permissions of the person asking.
National ID (TCKN), IBAN, card number and email are masked before they reach the model. Field-level rules define which field is visible to which role.
After reading external content, the agent does not send free text without approval. Instructions hidden in an email or a customer question have no effect.
Reading is open, writing is controlled. You choose one of four tiers for each connector; the agent never works with a permission you have not defined.
The agent only reads, analyzes and reports.
It fully prepares and shows the record it would write, without touching the system. Ideal before taking a new task live.
It writes once the task owner approves. Approval is given in the app, by mobile notification, or over Slack, Teams or WhatsApp.
It writes on its own, within the limits you define.
Ready-made agents come in approved mode. Destructive operations such as deletion always require approval, whatever the tier.
Intent checking and a topic filter tuned to your company profile turn away off-topic questions without answering them; corporate tokens are not spent on off-topic chats. Blocked conversations are reported, so you see which topics were asked about. Bulk data export is also capped by a threshold.
Every step taken by agents and users goes into the audit log: who asked, which data was read, which action was proposed, who approved or rejected it. In an audit or a dispute, you go back and review it step by step.
Every model call, tool use and error is logged too. You export the records to your own monitoring system with OpenTelemetry and track issues with the tools you already use.
On the cost tracking screen you see token consumption and cost per user, agent and task. You know how much each agent spends and which task is expensive, and you adjust your model choice accordingly.
You also see separately how much cost each feature brings. You switch off a feature that does not create value with one click, and allow only the features that pay off.
You can start using KuikAgents in the cloud right away, or install it on your own infrastructure. If data must never leave the organization under any circumstances, everything runs in-house, including the model.
No setup project. Works with your own model key; masking in the Advanced and Enterprise packages.
Runs on your own infrastructure with a setup wizard that runs in the browser.
The model runs on your infrastructure too; no request leaves the organization.
KuikAgents services are being installed on your server with Docker. This step takes a few minutes.