The tool catalogue
The tools an agent can be given, each with a risk class. Tools run in Rykon AI’s control plane, never in the sandbox.
Updated 1 min read
Each tool in the catalogue declares what kind of action it is, and that decides its risk class, from R0 (read and report) to R4 (irreversible, outbound to strangers, money, deletion). When an agent calls a tool, Rykon AI decides by the agent’s level whether it runs at once or waits for a person.
| Tool | What it does | Class |
|---|---|---|
record_finding | Records one finding in the run’s record, up to 2,000 characters. | R0 |
list_findings | Lists the findings the run has recorded so far. | R0 |
Tools that reach the world, such as web fetch through an allowlist, your own MCP servers and Rykon Mail inboxes, each carry their own class and appear in the catalogue when your account connects them.
Why tools run outside the sandbox
The agent only asks. The control plane checks the call against the agent’s tools and level, records it, and runs it only if it is allowed or approved. So an agent cannot run a tool it was not given, and cannot get round a decision by anything it does inside its sandbox.