instructeer

Python client library for the Instructeer API.

Instructeer provides deterministic guardrails for LLM pipelines — no ML inference, no probabilistic scores. The same input always produces the same output.

• SQLGuard — AST-based SQL risk analysis. Blocks dangerous or expensive queries before they reach your database.
• PIIGuard — Regex and lexicon-based PII detection. Prevents sensitive data from leaking through prompts or logs.

Quick Start · Documentation

Installation

pip install instructeer

Requires Python 3.11+.

Authentication

All requests require an API key. Get yours at instructeer.com.

Pass the key directly or set the INSTRUCTEER_API_KEY environment variable:

# macOS / Linux
export INSTRUCTEER_API_KEY=rg_...

# Windows (cmd)
set INSTRUCTEER_API_KEY=rg_...

# Windows (PowerShell)
$env:INSTRUCTEER_API_KEY="rg_..."

SQLGuard

from instructeer.guards import SQLGuard

# api_key = os.environ["INSTRUCTEER_API_KEY"]
sql = SQLGuard(api_key="rg_...")
result = sql.analyze("SELECT * FROM users WHERE id = 1")

print(result.allowed)    # True
print(result.severity)   # "low"
print(result.issues)     # []

{ "allowed": true, "severity": "low", "issues": [] }

Block a query based on the result:

generated_sql = "DELETE FROM users"
result = sql.analyze(generated_sql)

# Block on high severity only (allowed is False)
if not result.allowed:
    raise ValueError(f"Query blocked: {result.issues[0].message}")

# Block on medium and high severity
if result.severity in ("medium", "high"):
    raise ValueError(f"Query blocked: {result.issues[0].message}")

{
  "allowed": false,
  "severity": "high",
  "issues": [
    { "rule": "dml_no_where", "severity": "high", "message": "DELETE without WHERE clause would affect all rows" }
  ]
}

Dialects

Defaults to "postgres". Supported values:

postgres · mysql · bigquery · tsql · oracle · snowflake · databricks · sqlite · hive · mariadb

result = sql.analyze("SELECT TOP 10 * FROM users", dialect="tsql")

Response

Field	Type	Description
allowed	bool	False when a high-severity issue is detected
severity	str	"low", "medium", or "high"
issues	list[Issue]	Each issue has rule, message, and severity

PIIGuard

from instructeer.guards import PIIGuard

# api_key = os.environ["INSTRUCTEER_API_KEY"]
pii = PIIGuard(api_key="rg_...")
result = pii.detect_all("Alice's SSN is 123-45-6789 and email is alice@example.com")

print(result.summary.total)           # 2
print(result.summary.max_severity)    # "high"
for d in result.detections:
    print(d.entity_type, d.value, d.confidence)

{
  "detections": [
    { "entity_type": "US_SSN",        "value": "123-45-6789",        "confidence": 1.0, "severity": "high" },
    { "entity_type": "EMAIL_ADDRESS", "value": "alice@example.com",  "confidence": 1.0, "severity": "medium" }
  ],
  "summary": {
    "total": 2,
    "max_severity": "high",
    "by_type": { "US_SSN": 1, "EMAIL_ADDRESS": 1 },
    "by_severity": { "high": 1, "medium": 1 }
  }
}

Single detectors

result = pii.detect_email("Reach out at hello@example.com")
result = pii.detect_ssn("Patient SSN: 078-05-1120")
result = pii.detect_card("Charged to 4111 1111 1111 1111")
result = pii.detect_phone("Call 800-555-0199")

Available: detect_email · detect_phone · detect_ssn · detect_card · detect_bank · detect_iban · detect_ip · detect_url · detect_date · detect_dob · detect_dl · detect_passport · detect_address · detect_person

Filtering

result = pii.detect_all(
    text,
    detectors=["email", "ssn", "card"],
    min_confidence=0.8,
    confidence_thresholds={"US_ADDRESS": 0.7},
    use_default_thresholds=True,
)

Response

DetectResponse

Field	Type	Description
detections	list[Detection]	All detected PII spans
summary	DetectionSummary	Aggregate counts and severity roll-up

Detection

Field	Type	Description
entity_type	str	e.g. "EMAIL_ADDRESS", "US_SSN", "CREDIT_CARD"
value	str	The matched text
start	int	Start character offset
end	int	End character offset
severity	str	"low", "medium", or "high"
confidence	float	Detection confidence (0.0–1.0)
extra	dict | None	Additional detector-specific metadata

DetectionSummary

Field	Type	Description
total	int	Total number of detections
max_severity	str | None	Highest severity with confidence ≥ 0.3
by_type	dict[str, int]	Count per entity type
by_severity	dict[str, int]	Count per severity level

Async

import asyncio
from instructeer.guards import AsyncSQLGuard, AsyncPIIGuard

async def main():
    async with AsyncSQLGuard(api_key="rg_...") as sql:
        result = await sql.analyze("DROP TABLE users")

    async with AsyncPIIGuard(api_key="rg_...") as pii:
        result = await pii.detect_all("SSN: 123-45-6789")

asyncio.run(main())

Error handling

from instructeer import (
    AuthenticationError,
    PermissionDeniedError,
    RateLimitError,
    UnprocessableEntityError,
    APIConnectionError,
    APITimeoutError,
    APIError,
)

try:
    result = sql.analyze(query)
except AuthenticationError:
    ...
except PermissionDeniedError:
    ...
except RateLimitError as e:
    print(e.status_code)   # 429
except UnprocessableEntityError:
    ...
except APITimeoutError:
    ...
except APIConnectionError:
    ...
except APIError as e:
    print(e.status_code)

Configuration

Parameter	Default	Description
api_key	env INSTRUCTEER_API_KEY	Your Instructeer API key
base_url	https://api.instructeer.com	Override for self-hosted deployments
timeout	30.0	Request timeout in seconds
max_retries	2	Retries on 429 and 5xx with exponential backoff

License

MIT