What you get
On top of Claude Code
The working model stays familiar: same tools, same approach to context. On top
of it sits a layer that splits work across several agents and merges the result.
Orchestration
Tasks and subtasks, assignment to agents, statuses, handover down the chain.
Work is not lost between runs.
Extensibility
Your own agents, skills and integrations. External services connect over MCP, so
an agent reaches your API without platform changes.
Control
Run logs, per-agent permissions, manual confirmation for irreversible steps. You
can see who did what and on what grounds.
How it works
A single agent hits a ceiling once the task stops fitting into one context. From
there it turns manual: split it up, run the pieces, collect the output. Rosary OS
takes that mechanics over. You describe roles and rules, the platform keeps the
queue and the state.
Extension goes through configuration and MCP. An agent is handed a toolset,
skills come from the shared catalogue or are written for the project. Permissions
are scoped to the role, so an agent with production database access exists only
where you deliberately allowed it.
Example scenario
A large task is split across several agents: one prepares the change, another
checks it against its own rules, a third assembles the report. The irreversible
step waits for a human. This illustrates how a process is structured; it is not a
description of a deployment at a named company.
Where to go next
- Documentation: details and reference.
- How the platform works: the Rosary OS overview.
- Pricing: limits and terms.