Skip to main content
Version: dev

Concepts

This section explains how Rossoctl works. It has two parts.

Core describes the parts that every Rossoctl installation uses. These parts are ready for production use. Read them in the order that the sidebar shows.

Experiments describes features that are not yet ready for production use. Each experiment page gives the status of the feature and tells you how to enable it. You can ignore this group. A Rossoctl installation without any experiment is complete and supported.

The split follows the Rossoctl organization, which marks each repository as core or experimental.

Maturity levels

LevelWhat it means
ReadyThe feature is on by default. Tests cover it. You can depend on it.
BetaThe feature works and this site documents it. The configuration can change in a later release.
AlphaYou can evaluate the feature. The behaviour and the interface will change. Do not depend on it.

Core features

FeatureStatusPage
Agent identityReadyIdentity and trust
Access control and delegationReadyIdentity and trust
Agent deployment and discoveryReadyControl plane
Visibility of each model, tool and API callReadyRossoCortex
Traces and network dataReadyObservability
Token counts, cost and latency on a laptopReadyRead the numbers

Experimental features

FeatureStatusWhat it doesPage
MCP GatewayBetaGives every agent one address for all tools.MCP Gateway
SkillsBetaStores reusable instructions that an agent can use.Skills
Agent contextBetaGives an agent durable storage for files and memory.Agent context
SandboxesAlphaIsolates an agent more strongly than a pod does.Sandboxes
Intent-based accessAlphaDenies an action that does not match the request of the user.Intent-based access
Tool call validationAlphaDenies a tool call that has invented arguments.Tool call validation
Context compactionAlphaMakes tool output smaller before the model reads it.Context compaction
Cost controlAlphaRemoves unused tool definitions and limits spending.Cost control

Two more experiments exist in the platform but have no page on this site yet. Failure recovery returns a stopped agent to a known state. Data-flow analysis records the source of the data that an agent used. For the current state of both, see the cortex repository.

Report an error

If a feature does not operate as this page describes, the error is in these documents. Open an issue on rossoctl/rossoctl, or write a message in Slack.