mirror of
https://github.com/BerriAI/litellm.git
synced 2026-11-12 08:42:18 +00:00
style:fix ruff lint, ruff format, and type-discipline gate violations
This commit is contained in:
AlphaRex-pixel
parent
b7ecab516a
commit
b30027b229
4 changed files with 253 additions and 179 deletions
|
|
@ -1,3 +1,4 @@
|
|||
from types import MappingProxyType
|
||||
from typing import TYPE_CHECKING, Final
|
||||
|
||||
from litellm.types.guardrails import SupportedGuardrailIntegrations
|
||||
|
|
@ -23,10 +24,14 @@ def initialize_guardrail(litellm_params: "LitellmParams", guardrail: "Guardrail"
|
|||
return _tealtiger_callback
|
||||
|
||||
|
||||
guardrail_initializer_registry: Final = {
|
||||
SupportedGuardrailIntegrations.TEALTIGER.value: initialize_guardrail,
|
||||
}
|
||||
guardrail_initializer_registry: Final = MappingProxyType(
|
||||
{
|
||||
SupportedGuardrailIntegrations.TEALTIGER.value: initialize_guardrail,
|
||||
}
|
||||
)
|
||||
|
||||
guardrail_class_registry: Final = {
|
||||
SupportedGuardrailIntegrations.TEALTIGER.value: TealTigerGuardrail,
|
||||
}
|
||||
guardrail_class_registry: Final = MappingProxyType(
|
||||
{
|
||||
SupportedGuardrailIntegrations.TEALTIGER.value: TealTigerGuardrail,
|
||||
}
|
||||
)
|
||||
|
|
|
|||
|
|
@ -7,9 +7,12 @@ from __future__ import annotations
|
|||
|
||||
import threading
|
||||
import uuid
|
||||
from collections.abc import Mapping, Sequence
|
||||
from dataclasses import dataclass, field
|
||||
from datetime import datetime
|
||||
from datetime import date
|
||||
from enum import Enum
|
||||
from types import MappingProxyType
|
||||
from typing import Final
|
||||
|
||||
from .patterns import default_patterns
|
||||
|
||||
|
|
@ -25,51 +28,61 @@ class Action(str, Enum):
|
|||
BLOCK = "BLOCK"
|
||||
|
||||
|
||||
@dataclass
|
||||
@dataclass(frozen=True)
|
||||
class Decision:
|
||||
action: str
|
||||
reason_code: str
|
||||
correlation_id: str
|
||||
redacted_text: str | None = None
|
||||
details: dict = field(default_factory=dict)
|
||||
details: Mapping[str, object] = field(default_factory=lambda: MappingProxyType({}))
|
||||
|
||||
def to_receipt(self) -> dict:
|
||||
return {
|
||||
"correlation_id": self.correlation_id,
|
||||
"action": self.action,
|
||||
"reason_code": self.reason_code,
|
||||
"details": self.details,
|
||||
}
|
||||
def to_receipt(self) -> Mapping[str, object]:
|
||||
return MappingProxyType(
|
||||
{
|
||||
"correlation_id": self.correlation_id,
|
||||
"action": self.action,
|
||||
"reason_code": self.reason_code,
|
||||
"details": self.details,
|
||||
}
|
||||
)
|
||||
|
||||
|
||||
class TealEngine:
|
||||
def __init__(self, policies: list[dict], mode: PolicyMode = PolicyMode.ENFORCE) -> None:
|
||||
def __init__(
|
||||
self,
|
||||
policies: Sequence[Mapping[str, object]],
|
||||
mode: PolicyMode = PolicyMode.ENFORCE,
|
||||
) -> None:
|
||||
self.policies = policies
|
||||
self.mode = mode
|
||||
self.patterns = default_patterns()
|
||||
self._lock = threading.Lock()
|
||||
self._spend_by_day: dict[str, float] = {}
|
||||
self._spend_by_day: dict[str, float] = {} # mutable-ok: live counter under lock
|
||||
|
||||
self._pii_policy = next((p for p in policies if p["type"] == "pii"), None)
|
||||
self._cost_policy = next((p for p in policies if p["type"] == "cost"), None)
|
||||
self._tool_policy = next((p for p in policies if p["type"] == "tool_auth"), None)
|
||||
self._tool_policy = next(
|
||||
(p for p in policies if p["type"] == "tool_auth"), None
|
||||
)
|
||||
|
||||
def evaluate_text(self, text: str) -> Decision:
|
||||
correlation_id = str(uuid.uuid4())
|
||||
correlation_id: Final = str(uuid.uuid4())
|
||||
if not text or not self._pii_policy:
|
||||
return Decision(Action.ALLOW.value, "NO_VIOLATIONS", correlation_id)
|
||||
|
||||
findings = [name for name, pat in self.patterns.items() if pat.search(text)]
|
||||
findings: Final = tuple(
|
||||
name for name, pat in self.patterns.items() if pat.search(text)
|
||||
)
|
||||
if not findings:
|
||||
return Decision(Action.ALLOW.value, "NO_VIOLATIONS", correlation_id)
|
||||
|
||||
pii_action = self._pii_policy.get("action", "REDACT")
|
||||
pii_action: Final = self._pii_policy.get("action", "REDACT")
|
||||
if pii_action == "BLOCK" and self.mode == PolicyMode.ENFORCE:
|
||||
return Decision(
|
||||
Action.BLOCK.value,
|
||||
f"PII_DETECTED:{','.join(findings)}",
|
||||
correlation_id,
|
||||
details={"types": findings},
|
||||
details=MappingProxyType({"types": findings}),
|
||||
)
|
||||
|
||||
if self.mode == PolicyMode.MONITOR:
|
||||
|
|
@ -77,25 +90,29 @@ class TealEngine:
|
|||
Action.ALLOW.value,
|
||||
f"PII_DETECTED:{','.join(findings)}",
|
||||
correlation_id,
|
||||
details={"types": findings},
|
||||
details=MappingProxyType({"types": findings}),
|
||||
)
|
||||
|
||||
redacted = text
|
||||
for name in findings:
|
||||
redacted = self.patterns[name].sub(f"[REDACTED:{name.upper()}]", redacted)
|
||||
redacted: Final = self._redact(text, findings)
|
||||
return Decision(
|
||||
Action.REDACT.value,
|
||||
f"PII_DETECTED:{','.join(findings)}",
|
||||
correlation_id,
|
||||
redacted_text=redacted,
|
||||
details={"types": findings},
|
||||
details=MappingProxyType({"types": findings}),
|
||||
)
|
||||
|
||||
def _redact(self, text: str, findings: Sequence[str]) -> str:
|
||||
result = text # rebind-ok: progressively transformed across findings, one sub() pass per match
|
||||
for name in findings:
|
||||
result = self.patterns[name].sub(f"[REDACTED:{name.upper()}]", result)
|
||||
return result
|
||||
|
||||
def check_tool(self, tool_name: str) -> bool:
|
||||
if not self._tool_policy:
|
||||
return True
|
||||
allowlist = self._tool_policy.get("allowlist")
|
||||
blocklist = self._tool_policy.get("blocklist")
|
||||
allowlist: Final = self._tool_policy.get("allowlist")
|
||||
blocklist: Final = self._tool_policy.get("blocklist")
|
||||
if blocklist and tool_name in blocklist:
|
||||
return False
|
||||
if allowlist is not None:
|
||||
|
|
@ -105,17 +122,17 @@ class TealEngine:
|
|||
def check_budget(self, session_id: str = "default") -> tuple[bool, float, float]:
|
||||
if not self._cost_policy:
|
||||
return False, 0.0, 0.0
|
||||
limit = self._cost_policy.get("daily_limit_usd")
|
||||
limit: Final = self._cost_policy.get("daily_limit_usd")
|
||||
if limit is None:
|
||||
return False, 0.0, 0.0
|
||||
today = datetime.now().astimezone().date().isoformat()
|
||||
today: Final = date.today().isoformat() # noqa: DTZ011 # daily budget boundary is intentionally local-server-time, not UTC
|
||||
with self._lock:
|
||||
spent = self._spend_by_day.get(today, 0.0)
|
||||
spent: Final = self._spend_by_day.get(today, 0.0)
|
||||
return spent >= limit, spent, limit
|
||||
|
||||
def track_cost(self, tokens: int, cost_per_1k_tokens: float = 0.002) -> float:
|
||||
cost = (tokens / 1000.0) * cost_per_1k_tokens
|
||||
today = datetime.now().astimezone().date().isoformat()
|
||||
cost: Final = (tokens / 1000.0) * cost_per_1k_tokens
|
||||
today: Final = date.today().isoformat() # noqa: DTZ011 # daily budget boundary is intentionally local-server-time, not UTC
|
||||
with self._lock:
|
||||
self._spend_by_day[today] = self._spend_by_day.get(today, 0.0) + cost
|
||||
return cost
|
||||
|
|
|
|||
|
|
@ -1,7 +1,7 @@
|
|||
"""
|
||||
Regex-based PII / secrets detection patterns for TealTiger.
|
||||
|
||||
40+ patterns across five categories: government IDs, financial data,
|
||||
47 patterns across five categories: government IDs, financial data,
|
||||
contact info, network/device identifiers, and credentials/API keys.
|
||||
|
||||
These are deterministic regex matches — fast (no ML, no network calls) but
|
||||
|
|
@ -11,125 +11,154 @@ Callers can narrow scope via TealEngine(policies=[{"patterns": [...]}])
|
|||
instead of "all", or override this module's PII_PATTERNS entirely.
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import re
|
||||
from collections.abc import Mapping
|
||||
from types import MappingProxyType
|
||||
from typing import Final
|
||||
|
||||
PII_PATTERNS = {
|
||||
# ---------- government / national IDs ----------
|
||||
"ssn": re.compile(r"\b\d{3}-\d{2}-\d{4}\b"),
|
||||
"ssn_no_dash": re.compile(r"\b(?!000|666|9\d{2})\d{3}(?!00)\d{2}(?!0000)\d{4}\b"),
|
||||
"itin": re.compile(r"\b9\d{2}-(7\d|8[0-8])-\d{4}\b"),
|
||||
"ein": re.compile(r"\b\d{2}-\d{7}\b"),
|
||||
"us_passport": re.compile(r"\b[A-Z]{1,2}\d{7,9}\b"),
|
||||
"passport_generic_intl": re.compile(r"\b[A-Z]{2}\d{6,9}\b"),
|
||||
"drivers_license_generic": re.compile(r"\b[A-Z]{1,2}\d{6,8}\b"),
|
||||
"medicare_id_us": re.compile(r"\b\d[A-Z]\d{2}-[A-Z]\d{2}-[A-Z]\d{2}\b"),
|
||||
"npi_number": re.compile(r"\b\d{10}\b"),
|
||||
"uk_nino": re.compile(r"\b[A-CEGHJ-PR-TW-Z]{2}\d{6}[A-D]\b"),
|
||||
"date_of_birth": re.compile(r"\b(0[1-9]|1[0-2])[/-](0[1-9]|[12]\d|3[01])[/-](19|20)\d{2}\b"),
|
||||
"vin": re.compile(r"\b[A-HJ-NPR-Z0-9]{17}\b"),
|
||||
# ---------- financial ----------
|
||||
"credit_card_visa": re.compile(r"\b4\d{3}[ -]?\d{4}[ -]?\d{4}[ -]?\d{4}\b"),
|
||||
"credit_card_mastercard": re.compile(r"\b5[1-5]\d{2}[ -]?\d{4}[ -]?\d{4}[ -]?\d{4}\b"),
|
||||
"credit_card_amex": re.compile(r"\b3[47]\d{2}[ -]?\d{6}[ -]?\d{5}\b"),
|
||||
"credit_card_discover": re.compile(r"\b6(?:011|5\d{2})[ -]?\d{4}[ -]?\d{4}[ -]?\d{4}\b"),
|
||||
"credit_card_generic": re.compile(r"\b(?:\d[ -]*?){13,16}\b"),
|
||||
"iban": re.compile(r"\b[A-Z]{2}\d{2}[A-Z0-9]{10,30}\b"),
|
||||
"swift_bic": re.compile(r"\b[A-Z]{4}[A-Z]{2}[A-Z0-9]{2}(?:[A-Z0-9]{3})?\b"),
|
||||
"us_bank_routing": re.compile(r"\b\d{9}\b"),
|
||||
"us_bank_account": re.compile(r"\b\d{8,17}\b"),
|
||||
"bitcoin_address": re.compile(r"\b(bc1|[13])[a-zA-HJ-NP-Z0-9]{25,39}\b"),
|
||||
"ethereum_address": re.compile(r"\b0x[a-fA-F0-9]{40}\b"),
|
||||
# ---------- contact info ----------
|
||||
"email": re.compile(r"\b[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Za-z]{2,}\b"),
|
||||
"phone_us": re.compile(r"\b(?:\+?1[-.\s]?)?\(?\d{3}\)?[-.\s]?\d{3}[-.\s]?\d{4}\b"),
|
||||
"phone_intl": re.compile(r"\+\d{1,3}[-.\s]?\(?\d{1,4}\)?(?:[-.\s]?\d{2,4}){2,4}\b"),
|
||||
"us_zip_plus4": re.compile(r"\b\d{5}-\d{4}\b"),
|
||||
"po_box": re.compile(r"\bP\.?O\.?\s?Box\s?\d+\b", re.IGNORECASE),
|
||||
"lat_long_coordinates": re.compile(r"\b-?\d{1,3}\.\d{3,},\s?-?\d{1,3}\.\d{3,}\b"),
|
||||
# ---------- network / device identifiers ----------
|
||||
"ipv4": re.compile(r"\b(?:\d{1,3}\.){3}\d{1,3}\b"),
|
||||
"ipv6": re.compile(r"\b(?:[A-Fa-f0-9]{1,4}:){7}[A-Fa-f0-9]{1,4}\b"),
|
||||
"mac_address": re.compile(r"\b(?:[0-9A-Fa-f]{2}:){5}[0-9A-Fa-f]{2}\b"),
|
||||
"imei": re.compile(r"\b\d{15}\b"),
|
||||
"url_with_credentials": re.compile(r"\bhttps?://[^:\s]+:[^@\s]+@[^\s]+\b"),
|
||||
# ---------- credentials / API keys / secrets ----------
|
||||
"aws_access_key": re.compile(r"\b(AKIA|ASIA)[0-9A-Z]{16}\b"),
|
||||
"aws_secret_key": re.compile(r"\b(?=.*[A-Za-z])(?=.*\d)(?=.*[+/])[A-Za-z0-9+/]{40}\b"),
|
||||
"gcp_api_key": re.compile(r"\bAIza[0-9A-Za-z_-]{35}\b"),
|
||||
"azure_client_secret": re.compile(r"\b[A-Za-z0-9_~.-]{3}\dQ~[A-Za-z0-9_~.-]{31,34}\b"),
|
||||
"github_token": re.compile(r"\bgh[pousr]_[A-Za-z0-9]{36,255}\b"),
|
||||
"gitlab_token": re.compile(r"\bglpat-[A-Za-z0-9_-]{20}\b"),
|
||||
"slack_token": re.compile(r"\bxox[baprs]-[A-Za-z0-9-]{10,72}\b"),
|
||||
"stripe_live_key": re.compile(r"\b(sk|pk)_live_[A-Za-z0-9]{24,}\b"),
|
||||
"openai_api_key": re.compile(r"\bsk-[A-Za-z0-9]{20,}\b"),
|
||||
"jwt": re.compile(r"\beyJ[A-Za-z0-9_-]+\.[A-Za-z0-9_-]+\.[A-Za-z0-9_-]+\b"),
|
||||
"private_key_pem": re.compile(r"-----BEGIN (RSA |EC |OPENSSH |)PRIVATE KEY-----"),
|
||||
"generic_bearer_token": re.compile(r"\bBearer\s+[A-Za-z0-9_\-.]{20,}\b"),
|
||||
"basic_auth_header": re.compile(r"\bBasic\s+[A-Za-z0-9+/]{20,}={0,2}\b"),
|
||||
}
|
||||
PII_PATTERNS: Final = MappingProxyType(
|
||||
{
|
||||
# ---------- government / national IDs ----------
|
||||
"ssn": re.compile(r"\b\d{3}-\d{2}-\d{4}\b"),
|
||||
"ssn_no_dash": re.compile(
|
||||
r"\b(?!000|666|9\d{2})\d{3}(?!00)\d{2}(?!0000)\d{4}\b"
|
||||
),
|
||||
"itin": re.compile(r"\b9\d{2}-(7\d|8[0-8])-\d{4}\b"),
|
||||
"ein": re.compile(r"\b\d{2}-\d{7}\b"),
|
||||
"us_passport": re.compile(r"\b[A-Z]{1,2}\d{7,9}\b"),
|
||||
"passport_generic_intl": re.compile(r"\b[A-Z]{2}\d{6,9}\b"),
|
||||
"drivers_license_generic": re.compile(r"\b[A-Z]{1,2}\d{6,8}\b"),
|
||||
"medicare_id_us": re.compile(r"\b\d[A-Z]\d{2}-[A-Z]\d{2}-[A-Z]\d{2}\b"),
|
||||
"npi_number": re.compile(r"\b\d{10}\b"),
|
||||
"uk_nino": re.compile(r"\b[A-CEGHJ-PR-TW-Z]{2}\d{6}[A-D]\b"),
|
||||
"date_of_birth": re.compile(
|
||||
r"\b(0[1-9]|1[0-2])[/-](0[1-9]|[12]\d|3[01])[/-](19|20)\d{2}\b"
|
||||
),
|
||||
"vin": re.compile(r"\b[A-HJ-NPR-Z0-9]{17}\b"),
|
||||
# ---------- financial ----------
|
||||
"credit_card_visa": re.compile(r"\b4\d{3}[ -]?\d{4}[ -]?\d{4}[ -]?\d{4}\b"),
|
||||
"credit_card_mastercard": re.compile(
|
||||
r"\b5[1-5]\d{2}[ -]?\d{4}[ -]?\d{4}[ -]?\d{4}\b"
|
||||
),
|
||||
"credit_card_amex": re.compile(r"\b3[47]\d{2}[ -]?\d{6}[ -]?\d{5}\b"),
|
||||
"credit_card_discover": re.compile(
|
||||
r"\b6(?:011|5\d{2})[ -]?\d{4}[ -]?\d{4}[ -]?\d{4}\b"
|
||||
),
|
||||
"credit_card_generic": re.compile(r"\b(?:\d[ -]*?){13,16}\b"),
|
||||
"iban": re.compile(r"\b[A-Z]{2}\d{2}[A-Z0-9]{10,30}\b"),
|
||||
"swift_bic": re.compile(r"\b[A-Z]{4}[A-Z]{2}[A-Z0-9]{2}(?:[A-Z0-9]{3})?\b"),
|
||||
"us_bank_routing": re.compile(r"\b\d{9}\b"),
|
||||
"us_bank_account": re.compile(r"\b\d{8,17}\b"),
|
||||
"bitcoin_address": re.compile(r"\b(bc1|[13])[a-zA-HJ-NP-Z0-9]{25,39}\b"),
|
||||
"ethereum_address": re.compile(r"\b0x[a-fA-F0-9]{40}\b"),
|
||||
# ---------- contact info ----------
|
||||
"email": re.compile(r"\b[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Za-z]{2,}\b"),
|
||||
"phone_us": re.compile(
|
||||
r"\b(?:\+?1[-.\s]?)?\(?\d{3}\)?[-.\s]?\d{3}[-.\s]?\d{4}\b"
|
||||
),
|
||||
"phone_intl": re.compile(
|
||||
r"\+\d{1,3}[-.\s]?\(?\d{1,4}\)?(?:[-.\s]?\d{2,4}){2,4}\b"
|
||||
),
|
||||
"us_zip_plus4": re.compile(r"\b\d{5}-\d{4}\b"),
|
||||
"po_box": re.compile(r"\bP\.?O\.?\s?Box\s?\d+\b", re.IGNORECASE),
|
||||
"lat_long_coordinates": re.compile(
|
||||
r"\b-?\d{1,3}\.\d{3,},\s?-?\d{1,3}\.\d{3,}\b"
|
||||
),
|
||||
# ---------- network / device identifiers ----------
|
||||
"ipv4": re.compile(r"\b(?:\d{1,3}\.){3}\d{1,3}\b"),
|
||||
"ipv6": re.compile(r"\b(?:[A-Fa-f0-9]{1,4}:){7}[A-Fa-f0-9]{1,4}\b"),
|
||||
"mac_address": re.compile(r"\b(?:[0-9A-Fa-f]{2}:){5}[0-9A-Fa-f]{2}\b"),
|
||||
"imei": re.compile(r"\b\d{15}\b"),
|
||||
"url_with_credentials": re.compile(r"\bhttps?://[^:\s]+:[^@\s]+@[^\s]+\b"),
|
||||
# ---------- credentials / API keys / secrets ----------
|
||||
"aws_access_key": re.compile(r"\b(AKIA|ASIA)[0-9A-Z]{16}\b"),
|
||||
"aws_secret_key": re.compile(
|
||||
r"\b(?=.*[A-Za-z])(?=.*\d)(?=.*[+/])[A-Za-z0-9+/]{40}\b"
|
||||
),
|
||||
"gcp_api_key": re.compile(r"\bAIza[0-9A-Za-z_-]{35}\b"),
|
||||
"azure_client_secret": re.compile(
|
||||
r"\b[A-Za-z0-9_~.-]{3}\dQ~[A-Za-z0-9_~.-]{31,34}\b"
|
||||
),
|
||||
"github_token": re.compile(r"\bgh[pousr]_[A-Za-z0-9]{36,255}\b"),
|
||||
"gitlab_token": re.compile(r"\bglpat-[A-Za-z0-9_-]{20}\b"),
|
||||
"slack_token": re.compile(r"\bxox[baprs]-[A-Za-z0-9-]{10,72}\b"),
|
||||
"stripe_live_key": re.compile(r"\b(sk|pk)_live_[A-Za-z0-9]{24,}\b"),
|
||||
"openai_api_key": re.compile(r"\bsk-[A-Za-z0-9]{20,}\b"),
|
||||
"jwt": re.compile(r"\beyJ[A-Za-z0-9_-]+\.[A-Za-z0-9_-]+\.[A-Za-z0-9_-]+\b"),
|
||||
"private_key_pem": re.compile(
|
||||
r"-----BEGIN (RSA |EC |OPENSSH |)PRIVATE KEY-----"
|
||||
),
|
||||
"generic_bearer_token": re.compile(r"\bBearer\s+[A-Za-z0-9_\-.]{20,}\b"),
|
||||
"basic_auth_header": re.compile(r"\bBasic\s+[A-Za-z0-9+/]{20,}={0,2}\b"),
|
||||
}
|
||||
)
|
||||
|
||||
|
||||
def default_patterns() -> dict:
|
||||
"""Return a fresh copy of the built-in pattern set."""
|
||||
return dict(PII_PATTERNS)
|
||||
def default_patterns() -> Mapping[str, re.Pattern[str]]:
|
||||
"""Return the built-in, immutable pattern mapping."""
|
||||
return PII_PATTERNS
|
||||
|
||||
|
||||
PATTERN_CATEGORIES = {
|
||||
"government_id": [
|
||||
"ssn",
|
||||
"ssn_no_dash",
|
||||
"itin",
|
||||
"ein",
|
||||
"us_passport",
|
||||
"passport_generic_intl",
|
||||
"drivers_license_generic",
|
||||
"medicare_id_us",
|
||||
"npi_number",
|
||||
"uk_nino",
|
||||
"date_of_birth",
|
||||
"vin",
|
||||
],
|
||||
"financial": [
|
||||
"credit_card_visa",
|
||||
"credit_card_mastercard",
|
||||
"credit_card_amex",
|
||||
"credit_card_discover",
|
||||
"credit_card_generic",
|
||||
"iban",
|
||||
"swift_bic",
|
||||
"us_bank_routing",
|
||||
"us_bank_account",
|
||||
"bitcoin_address",
|
||||
"ethereum_address",
|
||||
],
|
||||
"contact": [
|
||||
"email",
|
||||
"phone_us",
|
||||
"phone_intl",
|
||||
"us_zip_plus4",
|
||||
"po_box",
|
||||
"lat_long_coordinates",
|
||||
],
|
||||
"network_device": [
|
||||
"ipv4",
|
||||
"ipv6",
|
||||
"mac_address",
|
||||
"imei",
|
||||
"url_with_credentials",
|
||||
],
|
||||
"credentials": [
|
||||
"aws_access_key",
|
||||
"aws_secret_key",
|
||||
"gcp_api_key",
|
||||
"azure_client_secret",
|
||||
"github_token",
|
||||
"gitlab_token",
|
||||
"slack_token",
|
||||
"stripe_live_key",
|
||||
"openai_api_key",
|
||||
"jwt",
|
||||
"private_key_pem",
|
||||
"generic_bearer_token",
|
||||
"basic_auth_header",
|
||||
],
|
||||
}
|
||||
PATTERN_CATEGORIES: Final = MappingProxyType(
|
||||
{
|
||||
"government_id": (
|
||||
"ssn",
|
||||
"ssn_no_dash",
|
||||
"itin",
|
||||
"ein",
|
||||
"us_passport",
|
||||
"passport_generic_intl",
|
||||
"drivers_license_generic",
|
||||
"medicare_id_us",
|
||||
"npi_number",
|
||||
"uk_nino",
|
||||
"date_of_birth",
|
||||
"vin",
|
||||
),
|
||||
"financial": (
|
||||
"credit_card_visa",
|
||||
"credit_card_mastercard",
|
||||
"credit_card_amex",
|
||||
"credit_card_discover",
|
||||
"credit_card_generic",
|
||||
"iban",
|
||||
"swift_bic",
|
||||
"us_bank_routing",
|
||||
"us_bank_account",
|
||||
"bitcoin_address",
|
||||
"ethereum_address",
|
||||
),
|
||||
"contact": (
|
||||
"email",
|
||||
"phone_us",
|
||||
"phone_intl",
|
||||
"us_zip_plus4",
|
||||
"po_box",
|
||||
"lat_long_coordinates",
|
||||
),
|
||||
"network_device": (
|
||||
"ipv4",
|
||||
"ipv6",
|
||||
"mac_address",
|
||||
"imei",
|
||||
"url_with_credentials",
|
||||
),
|
||||
"credentials": (
|
||||
"aws_access_key",
|
||||
"aws_secret_key",
|
||||
"gcp_api_key",
|
||||
"azure_client_secret",
|
||||
"github_token",
|
||||
"gitlab_token",
|
||||
"slack_token",
|
||||
"stripe_live_key",
|
||||
"openai_api_key",
|
||||
"jwt",
|
||||
"private_key_pem",
|
||||
"generic_bearer_token",
|
||||
"basic_auth_header",
|
||||
),
|
||||
}
|
||||
)
|
||||
|
|
|
|||
|
|
@ -3,14 +3,18 @@ TealTigerGuardrail — self-contained PII / cost / tool-authorization guardrail.
|
|||
|
||||
Implements apply_guardrail(), the current recommended pattern for guardrails
|
||||
that need to run uniformly across chat completions, /v1/messages, responses
|
||||
API, embeddings, etc. Verified against a live litellm proxy (v1.96.0) and
|
||||
API, embeddings, etc. Verified against a live litellm proxy (v1.97.0) and
|
||||
against litellm/proxy/guardrails/guardrail_hooks/onyx/onyx.py as a reference
|
||||
implementation of the same interface.
|
||||
|
||||
No network calls, no LLM calls, no external API keys required.
|
||||
"""
|
||||
|
||||
from typing import TYPE_CHECKING, Final, Literal, Optional
|
||||
from __future__ import annotations
|
||||
|
||||
from collections.abc import Mapping, Sequence
|
||||
from types import MappingProxyType
|
||||
from typing import TYPE_CHECKING, Final, Literal
|
||||
|
||||
from litellm.integrations.custom_guardrail import (
|
||||
CustomGuardrail,
|
||||
|
|
@ -23,19 +27,23 @@ from .engine import Action, PolicyMode, TealEngine
|
|||
if TYPE_CHECKING:
|
||||
from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
|
||||
|
||||
DEFAULT_POLICIES = [
|
||||
{"type": "pii", "action": "REDACT", "patterns": "all"},
|
||||
{"type": "cost", "action": "ENFORCE", "daily_limit_usd": 50.0},
|
||||
{"type": "tool_auth", "action": "ENFORCE", "allowlist": None}, # None = allow all
|
||||
]
|
||||
_EMPTY_MAPPING: Final = MappingProxyType({})
|
||||
|
||||
DEFAULT_POLICIES: Final = (
|
||||
MappingProxyType({"type": "pii", "action": "REDACT", "patterns": "all"}),
|
||||
MappingProxyType({"type": "cost", "action": "ENFORCE", "daily_limit_usd": 50.0}),
|
||||
MappingProxyType(
|
||||
{"type": "tool_auth", "action": "ENFORCE", "allowlist": None}
|
||||
), # None = allow all
|
||||
)
|
||||
|
||||
|
||||
class TealTigerGuardrail(CustomGuardrail):
|
||||
def __init__(
|
||||
self,
|
||||
policies: list | None = None,
|
||||
policies: Sequence[Mapping[str, object]] | None = None,
|
||||
policy_mode: str = "ENFORCE",
|
||||
**kwargs: object,
|
||||
**kwargs, # kwargs-ok: forwards unknown/future CustomGuardrail.__init__ params (guardrail_name, event_hook, default_on, ...) to the base class
|
||||
) -> None:
|
||||
self.engine: Final = TealEngine(
|
||||
policies=policies or DEFAULT_POLICIES,
|
||||
|
|
@ -47,9 +55,9 @@ class TealTigerGuardrail(CustomGuardrail):
|
|||
async def apply_guardrail(
|
||||
self,
|
||||
inputs: GenericGuardrailAPIInputs,
|
||||
request_data: dict,
|
||||
request_data: Mapping[str, object],
|
||||
input_type: Literal["request", "response"],
|
||||
logging_obj: Optional["LiteLLMLoggingObj"] = None,
|
||||
logging_obj: LiteLLMLoggingObj | None = None,
|
||||
) -> GenericGuardrailAPIInputs:
|
||||
"""
|
||||
Runs on both request (pre-call) and response (post-call) content,
|
||||
|
|
@ -57,28 +65,43 @@ class TealTigerGuardrail(CustomGuardrail):
|
|||
|
||||
Tool-call authorization and budget checks only make sense on the
|
||||
request path, so they're gated on input_type == "request".
|
||||
|
||||
Returns a new mapping rather than mutating `inputs` in place, since
|
||||
the caller's object should not be rewritten at a distance.
|
||||
"""
|
||||
if input_type == "request":
|
||||
for tool_call in inputs.get("tool_calls") or []:
|
||||
tool_name = tool_call.get("name") or tool_call.get("function", {}).get("name")
|
||||
if tool_name and not self.engine.check_tool(tool_name):
|
||||
raise ValueError(f"TealTiger: blocked — TOOL_NOT_ALLOWLISTED ({tool_name})")
|
||||
self._check_tool_calls(inputs)
|
||||
self._check_budget(request_data)
|
||||
|
||||
over_budget, spent, limit = self.engine.check_budget(session_id=request_data.get("user") or "default")
|
||||
if over_budget:
|
||||
raise ValueError(f"TealTiger: blocked — DAILY_BUDGET_EXCEEDED (${spent:.2f} / ${limit:.2f})")
|
||||
checked_texts: Final = tuple(
|
||||
self._check_text_or_raise(text) for text in inputs.get("texts") or ()
|
||||
)
|
||||
return {**inputs, "texts": list(checked_texts)} # mutable-ok: dict interop
|
||||
|
||||
checked_texts: list[str] = []
|
||||
for text in inputs.get("texts") or []:
|
||||
decision = self.engine.evaluate_text(text)
|
||||
def _check_tool_calls(self, inputs: GenericGuardrailAPIInputs) -> None:
|
||||
for tool_call in inputs.get("tool_calls") or ():
|
||||
tool_name = tool_call.get("name") or tool_call.get(
|
||||
"function", _EMPTY_MAPPING
|
||||
).get("name")
|
||||
if tool_name and not self.engine.check_tool(tool_name):
|
||||
raise ValueError(
|
||||
f"TealTiger: blocked — TOOL_NOT_ALLOWLISTED ({tool_name})"
|
||||
)
|
||||
|
||||
if decision.action == Action.BLOCK.value:
|
||||
raise ValueError(f"TealTiger: blocked — {decision.reason_code}")
|
||||
def _check_budget(self, request_data: Mapping[str, object]) -> None:
|
||||
over_budget, spent, limit = self.engine.check_budget(
|
||||
session_id=request_data.get("user") or "default"
|
||||
)
|
||||
if over_budget:
|
||||
raise ValueError(
|
||||
f"TealTiger: blocked — DAILY_BUDGET_EXCEEDED (${spent:.2f} / ${limit:.2f})"
|
||||
)
|
||||
|
||||
checked_texts.append(decision.redacted_text or text)
|
||||
|
||||
inputs["texts"] = checked_texts
|
||||
return inputs
|
||||
def _check_text_or_raise(self, text: str) -> str:
|
||||
decision: Final = self.engine.evaluate_text(text)
|
||||
if decision.action == Action.BLOCK.value:
|
||||
raise ValueError(f"TealTiger: blocked — {decision.reason_code}")
|
||||
return decision.redacted_text or text
|
||||
|
||||
# ---- cost tracking on successful response ----
|
||||
# NOTE for reviewers: apply_guardrail is only handed extracted `texts` /
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue