CoreLayer RTD

Connect your agent

Open RTD → Tool protection · v2 in the existing console. Organisation admins and security researchers can create a connection, copy its one-time credential, and run the generated commands on the customer’s server. Build the beta from this repository with go build -o corelayer-runtime-v2 ./cmd/runtime-v2. It is not part of the stable installer yet.

The agent connects outward to the console. Route actual HTTP or remote MCP tool calls through the agent; installing it near an AI does not automatically intercept calls. Tool endpoints and upstream credentials remain local.

Choose what the AI may do

• Monitor: observe and forward, including calls that would be blocked.
• Enforce: stop disallowed tool calls before execution.
• Set Allow or Block for each exact tool name, recognized-secret detection, and a shared calls-per-minute budget.
• Optionally block anomaly scores at a threshold from 50 to 100.

Read the anomaly score

Scores range from 0 to 100 and describe observed signals, not attack probabilities. Signals include bursts, tools absent from the baseline, and unusually large arguments. The first 30 successful, explicitly allowed, policy-clean calls form a frozen baseline. A zero score does not bypass tool permissions.

Understand policy status

Policy changes are pending until the agent validates and saves them, then acknowledges the revision. Offline agents retain their last policy. Revoking a console connection stops sync; it does not stop the agent or its tool calls.