Guardrails

Runtime helpers for provider-agnostic memory input guardrails.

MemoryInputGuardrail(*, engine, effective_phrases, model_name)

Run one phrase-matcher guardrail check over memory write input.

Initializes one memory-input guardrail runtime helper.

Parameters:

Name	Type	Description	Default
engine	Any	Built PhraseMatcherEngine instance.	required
effective_phrases	tuple[str, ...]	Final merged blocked phrase set.	required
model_name	str	Resolved spaCy model name.	required

effective_phrases property

Return the final merged phrase set.

Returns:

Type	Description
tuple[str, ...]	tuple[str, ...]: Final blocked phrase set.

model_name property

Return the resolved spaCy model name.

Returns:

Name	Type	Description
str	str	spaCy model name.

check_messages(messages) async

Evaluate one add request's messages before provider write.

Parameters:

Name	Type	Description	Default
messages	list[Message] | None	Caller messages for one add request.	required

Returns:

Name	Type	Description
MemoryInputGuardrailDecision	MemoryInputGuardrailDecision	Safe/unsafe decision with optional reason.

check_text(content) async

Evaluate one provider-neutral input string before provider write.

Parameters:

Name	Type	Description	Default
content	str	Provider-neutral input text prepared by one caller helper.	required

Returns:

Name	Type	Description
MemoryInputGuardrailDecision	MemoryInputGuardrailDecision	Safe/unsafe decision with optional reason.

allow_memory_add(*, guardrail, user_id, agent_id, messages) async

Evaluate whether one memory add request may continue.

Parameters:

Name	Type	Description	Default
guardrail	MemoryInputGuardrail | None	Configured guardrail helper, if any.	required
user_id	str	User identifier for the memory operation.	required
agent_id	str	Agent identifier for the memory operation.	required
messages	list[Message] | None	Messages to evaluate before add(...).	required

Returns:

Name	Type	Description
bool	bool	True when the add request may continue, else False.

allow_memory_update(*, guardrail, user_id, agent_id, new_content) async

Evaluate whether one memory update request may continue.

Parameters:

Name	Type	Description	Default
guardrail	MemoryInputGuardrail | None	Configured guardrail helper, if any.	required
user_id	str | None	User identifier for the memory operation.	required
agent_id	str | None	Agent identifier for the memory operation.	required
new_content	str | None	Replacement content supplied to update(...).	required

Returns:

Name	Type	Description
bool	bool	True when the update request may continue, else False.

build_memory_input_guardrail(config)

Build one optional memory-input guardrail from config or environment.

Parameters:

Name	Type	Description	Default
config	dict[str, Any] | MemoryManagerConfig | None	Memory-manager config input.	required

Returns:

Type	Description
MemoryInputGuardrail | None	MemoryInputGuardrail | None: Built guardrail helper, or None when disabled.

Raises:

Type	Description
RuntimeError	If one required dependency or spaCy model is unavailable.
ValueError	If one configured guardrail payload is invalid.