on any annual plan
Govern every prompt, model and AI agent
Valid until June 30, 2026. Not combinable with the annual billing discount.
A dedicated security layer that filters content, detects PII, enforces policies, and prevents data loss — before any prompt reaches an external provider. Real-time protection, compliance-ready audit trails.
Sound Familiar?
Without a dedicated security layer, your AI usage is a compliance incident waiting to happen
Employees unknowingly paste customer names, social security numbers, credit card details, and health records directly into LLM prompts every day.
Prompts and responses flow without any scanning. Harmful, biased, or off-policy content reaches users unchecked — and there's no record of what was generated.
Developers paste internal codebases, architecture diagrams, and trade secrets into external models for help. Once sent, that data is gone.
Your AI usage policy exists as a PDF nobody reads. There's no technical mechanism to enforce which data goes where, or which models are approved.
When regulators ask "what AI decisions were made and with what data?" — you have no answer. Every interaction is invisible to compliance teams.
Data breaches and policy violations are found after the damage is done — in quarterly audits, incident reports, or worse, in the press.
How It Works
Every prompt and response passes through Optscale AI's guardrails — scanned, filtered, and logged in real time.
The Security & Guardrails layer sits inside the Optscale AI Gateway. Before any prompt reaches an external model, it's scanned for PII, checked against your content policies, and validated for data sensitivity. Responses are filtered on the way back. Every interaction generates a compliance-ready audit record.
🛡 Real-time content filtering for both prompts and model responses
👤 Automatic PII detection and redaction before data reaches external providers
📜 Policy enforcement engine — define what data goes where, per team
🔔 Real-time violation alerting with configurable escalation rules
Incoming Prompt
"Pull the account summary for John Smith, SSN 482-91-0037 and forecast Q4 revenue from the internal_financials_v3 dataset."
↓
🛡 Optscale AI Security Layer
PII Detection • Content Filter • Policy Check • DLP Scan
✗ PII detected
✗ Internal dataset ref
✓ Content policy OK
✓ Model approved
↓
↓ redacted & sanitized
Cleaned Prompt Sent to Model
"Pull the account summary for [REDACTED] and forecast Q4 revenue from the [INTERNAL_REF_REMOVED]dataset."
Capabilities
Each layer works in real time at the gateway level — no code changes, no SDK integrations, no workflow disruption.
Configurable filters that scan both prompts and responses for inappropriate, harmful, or off-policy content in real time — before users ever see it.
✓ Bi-directional scanning of prompts and responses
✓ Custom policy rules per team or department
✓ Toxicity, bias, and off-topic detection
✓ Block, flag, or log-only enforcement modes
Automatically identifies and redacts personally identifiable information — names, SSNs, credit cards, health records — before it reaches any external model provider.
✓ 50+ PII entity types recognized
✓ Auto-redact or block with configurable thresholds
✓ Reversible tokenization for internal use
✓ GDPR, HIPAA, and PCI-DSS aware patterns
Define company-wide AI usage policies — which data goes where, which models are approved, and what outputs are acceptable. Enforce them automatically at the gateway.
✓ Declarative YAML-based policy definitions
✓ Model allowlists and blocklists per team
✓ Data classification-based routing rules
✓ Versioned policies with rollback support
Prevents sensitive corporate data, trade secrets, and proprietary code from leaking to external LLMs through real-time prompt monitoring and pattern matching.
✓ Source code and IP detection in prompts
✓ Financial data and projection recognition
✓ Internal document and dataset reference blocking
✓ Custom regex and keyword pattern rules
Configuration
Define your security guardrails in simple YAML. Optscale AI enforces them at the gateway level — no SDK changes required. Policies are versioned, auditable, and can be scoped per team, per model, or company-wide.
🔴🟡🟢 gateway-example.com
from openai import OpenAI
client = OpenAI( base_url="https://gateway.optscale.ai/v1", api_key="kf-your-api-key" )
response = client.chat.completions.create(
model="auto", # Gateway picks the best model
messages=[{ "role": "user", "content": "Summarize Q3 revenue trends" }],
max_tokens=1000)
print(response.choices[0].message.content)
# Model used: llama-3-8b (routed by gateway)
# Cost: $0.0003 (vs $0.012 with GPT-4)
# Latency: 340ms
Explore the Platform
Start free with up to 5 seats. Deploy the gateway in hours, see cost savings within weeks.
US Office:
1250 Borregas Avenue Sunnyvale, CA 94089, US
UK Office:
124 City Road, London, EC1V 2NX, United Kingdom