Return a copy of all audit entries.

* Estimate cost for running agent

Estimate query cost based on EXPLAIN and statistics.

Estimate token and cost savings from caching. Args: token_count: Number of tokens in the cached content

Estimate savings from OpenAI caching. Returns: Tuple of (tokens_saved, cost_saved_usd)

100% savings on local cache hits.

Evaluate alert rules against current metrics. Args: metrics: Current metrics to evaluate Re

Evaluate a response against all criteria. Returns: Dictionary with overall score, dimension scores, and pass/fai

Run the prompt defense evaluation.

* Evaluate a condition expression string against a context dictionary. * Supports: equality, inequality, numeric comparisons, `in` operator, * boole

* Evaluate all applicable policies for an agent action. * Returns a rich PolicyDecisionResult with matched rule, timing, etc.

* Evaluate a single rule against an action.

Simulate evaluating an agent version against golden test cases.

Async version (just wraps sync for rule-based).

Evaluate multiple prompts keyed by identifier. Args: prompts: Mapping of ``{identifier: prompt_text}``. Returns:

Return per-category metrics dict.

Evaluate a completed conversation using intent-specific metrics. This is the key method that implements the problem statemen

Convenience wrapper that reads diff stats from *repo_path*. Args: agent_id: Unique agent identifier. config: Scope co

Public access to the AGT ``input`` intervention point evaluation.

Public access to the AGT ``input`` intervention point evaluation. Falls back to the shared adapter-level :class:`ExecutionContext` wh

Public access to the AGT ``input`` intervention point evaluation.

Public access to the AGT ``input`` intervention point evaluation.

Public access to the AGT ``input`` intervention point evaluation.

Public access to the AGT ``input`` intervention point evaluation.

Public access to the AGT ``input`` intervention point evaluation.

Public access to the AGT ``input`` intervention point evaluation. Returns ``None`` when the bridge is disabled (no policy/runtime was

Public access to the AGT ``output`` intervention point evaluation.

AGT ``output`` evaluation for buffered OpenAI stream output.

Public access to the AGT ``output`` intervention point evaluation. Returns ``None`` when the bridge is disabled (no policy/runtime wa

Evaluate plugin manifests against a governance policy. Returns 0 if all manifests comply, 1 if any violate.

AGT ``pre_tool_call`` evaluation for an ADK tool invocation.

AGT ``pre_tool_call`` evaluation for a Mistral tool-call response.

AGT ``pre_tool_call`` evaluation for a CrewAI tool call.

AGT ``pre_tool_call`` evaluation for an AutoGen FunctionCall.

AGT ``pre_tool_call`` evaluation for a single tool invocation.

AGT ``pre_tool_call`` evaluation for a LangChain tool invocation.

AGT ``pre_tool_call`` evaluation for a Gemini function call.

AGT ``pre_tool_call`` evaluation for a PydanticAI tool invocation.

AGT ``pre_tool_call`` evaluation for a Bedrock action-group event.

Evaluate a sensor reading against cold chain governance policy.

Return the underlying synchronous evaluator.

evictOldest removes the entry with the fewest interactions (least active).

Third LLM call that reads critique and rewrites system instructions.

Generate improved system instructions based on critique.

Example 1: Creating a basic agent metadata manifest.

Example 1: Basic adapter usage with manual setup

Example 2: Using the convenience function for one-liner setup

Example 2: Agent marketplace discovery scenario.

Example 3: Composing multiple agents with compatible IO contracts.

Example 3: Custom tool name mapping

Example 4: Using callbacks to handle blocked actions

Example 4: Updating trust scores based on executions.

Example 5: Accessing audit trail and statistics

Example 5: Saving and loading manifests.

Demonstrate Agent Hibernation feature. Problem: Agents sitting idle in memory cost money or RAM. Solution: Serialize agent state to

Example: Rule-based alerting

Example: Detecting anomalous agent behavior

Example: Processing audio with transcription

Example showing audit logging

Basic usage example

Example: Behavioral analysis with context

Demonstrate using both hibernation and time-travel together. This shows how the two features complement each other: - Hibernation re

Example: Generating compliance reports

Example: Constitutional AI value alignment

Example 3: Constraint Graphs - Multi-Dimensional Context Context isn't flat; it's a graph. We define what data exists (Data Graph),

Example: Adding custom constitutional rules

Example: Distributed tracing

Example: EU AI Act compliance checking

Example: Using complete compliance suite

Example: Using complete multimodal suite

Example: Using complete ML safety suite

Example 5: Integrated Workflow - All Features Together Demonstrates how all features work together for comprehensive governance.

Example: Detecting jailbreak attempts

Example: Metrics export endpoint (for Prometheus scraping)