Trusera Sidecar Tool icon

Trusera Sidecar Tool

Intercepts ALL agent tool calls and enforces Cedar policies from the Trusera platform. Prompt-injection proof.

Trusera Sidecar Tool

AI-generated

Summary

Intercepts all agent tool calls and enforces Trusera Cedar policies with pattern-based prompt-injection and PII detection, optional AI brain mode evaluation, and configurable enforcement (log/warn/block).

Inputs

None.

Output shape

AiTool

The node returns a gate tool (trusera_policy_gate) that the AI agent must call before any other tool call. The tool requires tool_name, tool_args (JSON string), reasoning, contains_pii, and data_summary. Execution results in an approved, warning, or block response based on policy evaluation and the node's enforcement mode setting.

Examples

Example 1: Agent attempts to use Gmail send_email tool

Default settings: agentName set, enforcementMode=warn, policySource=platform, enablePiiDetection=true, enablePromptInjection=true, enableBrainMode=false. The gate tool requires tool_name='gmail_send_email', tool_args=JSON string of arguments, reasoning, contains_pii, data_summary. Policy evaluation runs PII/injection checks and platform Cedar policies. If approved, the tool executes; if violations, enforcement mode determines outcome (warn logs, block throws, log continues silently).

Links

Discussion