Feature
Agents as governed resources
Define AI workers with a kind, Runtime, model, instructions and allowed MCP servers, then Invoke them and audit each Agent run.
An Agent is a governed resource like an Application: it has an Owner user, an Owner group and a Mode, and new Agents default to 0o740. Its kind is nlp, data-scraping, api-interaction, orchestration or custom.
Invoking an Agent needs write access, not read, because it spends tokens and money. Public Agents can be found through Search but never invoked anonymously.
Each Invoke records an Agent run with its input, output, status and token usage. The invoke response is a handle; reading the Agent run is the source of truth. An Idempotency-Key header makes a repeated Invoke return the existing Agent run instead of starting another.
More features
All features →- 01
Application lifecycle
Create, configure and delete Applications with Quotas per Cluster, Namespaces, Ownership links and Extended resources.
- 02
Pipelines and Pipeline runs
Record the CI/CD definition of each Application and every execution of it, with Commit statuses beside them.
- 03
Deployment tracking
Follow each Deployment of an Application on a Cluster, with service type, DNS, Stages and Events.