feat: add custom guardrails implementation [AL-197] #920
Conversation
github-actions
bot
added
test:uipath-langchain
ctiliescuuipath
force-pushed
the
feat/custom_guardrails
branch
2 times, most recently
from
7f02b34 to
e8ce7af
Compare
ctiliescuuipath
force-pushed
the
feat/custom_guardrails
branch
2 times, most recently
from
dbdf973 to
fbf9514
Compare
ctiliescuuipath
force-pushed
the
feat/custom_guardrails
branch
from
fe4e4c7 to
89ce1b7
Compare
| IS_NOT_EMPTY = "isNotEmpty" | ||
| MATCHES_REGEX = "matchesRegex" | ||
| STARTS_WITH = "startsWith" | ||
| FUNC = "func" |
There was a problem hiding this comment.
Do we really need this operator type? I agree with the function for evaluation, but I challenge the FUNC operator type.
| except Exception: | ||
| # If function raises an exception, treat as failure | ||
| passed = False | ||
| elif rule.operator == NumberOperator.EQUALS: |
There was a problem hiding this comment.
I propose we only keep the func evaluation. It is more flexible and we can easily convert all the statis operators into funcs during the conversion (when we create the guardrails from their definitions).
I think we can even have one single function like:
def evaluate_rule( rule_func, input_data, output_data ) -> tuple[bool, str | None]
| return True, "All number rule validations passed" | ||
|
|
||
|
|
||
| def evaluate_boolean_rule( |
There was a problem hiding this comment.
Does it work to have a single function to evaluate word|number|boolean rules?
| ) | ||
|
|
||
| @traced("evaluate_pre_custom_guardrails", run_type="uipath") | ||
| def evaluate_pre_custom_guardrails( |
There was a problem hiding this comment.
Use singular -> evaluate_pre_custom_guardrail. Same for the below functions too.
|
|
||
| DOES_NOT_EQUAL = "doesNotEqual" | ||
| EQUALS = "equals" | ||
| FUNC = "func" |
There was a problem hiding this comment.
do we still need this?
| Field(discriminator="rule_type"), | ||
| ] | ||
|
|
||
| AgentRule = Annotated[ |
There was a problem hiding this comment.
All types prefixed by Agent should stay in agent.py file.
| model_config = ConfigDict(populate_by_name=True, extra="allow") | ||
|
|
||
|
|
||
| class LowCodeCustomGuardrail(BaseGuardrail): |
There was a problem hiding this comment.
We already have a AgentCustomGuardrails model. Why not updating that one with the new types?
| "Guardrail", | ||
| "BuiltInValidatorGuardrail", | ||
| "CustomGuardrail", | ||
| "LowCodeCustomGuardrail", |
There was a problem hiding this comment.
We shouldn't expose these here. They are agent related.
2 participants
No description provided.