Trusera Agent icon

Trusera Agent

AI Agent with built-in policy enforcement. Intercepts every tool call at the code level — prompt injection proof. Drop-in replacement for the standard AI Agent.

Trusera Agent

AI-generated

Summary

Execute a policy-enforced AI agent loop, evaluating each tool call against PII, prompt-injection, content-filter, and Cedar policies before execution, with enforcement mode determining block/warn/log behavior.

Inputs

  • agentName (required) — Name for this agent in the Trusera platform
  • text — The input prompt for the agent
  • systemMessage — System message that sets the agent behavior
  • enforcementMode — What happens when a tool call violates a policy
  • policySource — Cedar policy source
  • inlineCedarDsl — Cedar policy DSL for tool call evaluation
  • enablePiiDetection — Whether to scan tool arguments for PII
  • enablePromptInjection — Whether to detect prompt injection in tool arguments
  • enableBrainMode — Whether to use an LLM to evaluate complex policies contextually (adds ~1s per tool call)
  • brainApiKey — API key for the brain mode LLM (OpenAI-compatible)
  • brainBaseUrl — Base URL for the brain mode LLM API
  • brainModel — Model for AI-powered policy evaluation
  • maxIterations — Maximum number of agent reasoning iterations

Output shape

Per input item returns {json: {output: blockReason|finalOutput, blocked: boolean, intermediateSteps: [], _trusera: {agentName, enforcement, decision, reason, toolCallsTotal, toolDebug}}, pairedItem: {item: index}}.

If a tool call is blocked, output contains blockReason and blocked: true; otherwise output contains final LLM output, blocked: false, and intermediate steps with brainAnalysis when enabled.

Examples

Example 1: Blocked tool call

enforcementMode: block, any policy violation halts the agent loop and returns a POLICY VIOLATION error to the LLM.

Example 2: Allowed execution with brain mode

enableBrainMode: true adds an LLM-powered contextual policy evaluation step (~1s per tool call) before tool execution.

Links

Discussion