Security

Agents should inherit less authority than the people who supervise them.

Konteks is built around scoped context, revocable credentials, explicit approval, tenant boundaries, and evidence. We publish controls we can stand behind—not certification theatre.

01

Tenant boundaries

Konteks authorization and catalog access are tenant-scoped. Credentials and permissions should grant only the access needed for the intended workspace and action.

02

Secrets stay secret

Provider and connector credentials are kept in protected secret storage. Public APIs, MCP discovery, prompts, screenshots, and routine logs must not reveal raw secret material.

03

Human authority

Costed work is designed around an explicit proposal and approval boundary. High-impact mutations remain subject to permissions, policy, budget, and human confirmation.

04

Auditable delivery

Sessions, proposal versions, approvals, runs, validation, usage, and outcomes are correlated so teams can understand what happened and why.

05

Revocable access

MCP and connector credentials can be scoped, rotated, revoked, and replaced. Access should be removed promptly when a person, agent, or integration no longer needs it.

06

Responsible disclosure

If you believe you found a vulnerability, do not test against other tenants or access data that is not yours. Send the details privately to hello@konteks.io.