Oniva Agents runs AI workflows with human approvals and a tamper-evident audit log. oniva/ui is the open design system for the interfaces around them.
Products
Oniva Agents runs the work. oniva/ui is an open React design system with components for agent interfaces, and you can use it in your own product.
Build agent workflows on a visual canvas. Runs are checkpointed, supervised by people, and recorded in a tamper-evident audit log.
297 open-source React components for dashboards and agent interfaces, from charts and tokens to approvals and run logs.
Inside Oniva Agents
Three parts of the engine, each with the detail on the Agents page.
Every step is stored in Postgres, so a crashed run resumes from committed state and a retried tool call replays its stored result.
See the harnessSeveral agents research in parallel. The report cites numbered sources, and a lint blocks citations that point nowhere.
See Deep WorkRepeated trials on recorded inputs. CI fails on a regression against a committed baseline.
See the evalsHow we build
Approval gates, runs that resume after a crash, and a tamper-evident audit log with signed entries. A reviewer can trace what an agent did and why.
People supervise agents through an interface. Approvals, progress, and run logs are real components in oniva/ui, and the demo on the Agents page is built from them.
Read and audit the code. oniva/ui is MIT and Oniva Agents is source-available. Commercial licenses are available on request.
Licenses
oniva/ui is MIT licensed. Oniva Agents is source-available under BSL 1.1 and converts to Apache 2.0 in 2030.
Contact
Running agents in a regulated team? Tell us what is in your way.
Email usbala@oniva.ai