Guardrails
Prompt-injection blocking, PII masking, content moderation, banned keywords and secret detection, applied per key or per team before and after every model call.
Guardrails
Policies on every request and every response
Guardrails run inside the gateway, so every application gets the same protection without writing its own checks.
Input guardrails inspect the prompt before it reaches the model; output guardrails inspect the response before it reaches the user. Each key or team can have its own policy.
Why it matters
Protect customers, data and the organisation
Stop data leaks
Personal data and secrets are masked or blocked before they leave.
Resist misuse
Prompt-injection attempts are detected and blocked.
Consistent policy
One set of rules for every application, managed centrally.
Guardrails
What the gateway checks
Prompt-injection blocking
Detects and blocks attempts to override instructions or extract hidden data.
PII masking
A PII detection engine masks names, emails and phone numbers, and can be configured for Bangladesh identifiers such as national ID, account and mobile numbers.
Content moderation
Blocks or flags harmful or inappropriate content in prompts and responses.
Banned keywords
Blocks requests or responses that contain terms you define.
Secret detection
Stops API keys, passwords and tokens from being sent to models.
Per-key and per-team policies
Different guardrails for different applications and teams.
How it works
Before and after the model call
Before the call
Mask PII, detect secrets, block injection and banned keywords.
After the call
Moderate the response and mask any personal data it contains.
Logged decisions
Every block, mask and flag is recorded in the request log.
Policies
Guardrails and where they run
Prompt injection
- Before the call
- Yes
- After the call
- —
- Action
- Block
PII masking
- Before the call
- Yes
- After the call
- Yes
- Action
- Mask
Content moderation
- Before the call
- Yes
- After the call
- Yes
- Action
- Block or flag
Banned keywords
- Before the call
- Yes
- After the call
- Yes
- Action
- Block
Secret detection
- Before the call
- Yes
- After the call
- —
- Action
- Block or mask
Included
With every Comptech product
On premises or air-gapped
Runs in your own data centre, so prompts and responses stay under your control.
Single sign-on
SAML and OIDC sign-in, with AD/LDAP groups mapped to teams and roles.
Secrets vault
Provider credentials held centrally, never handed to applications.
DC/DR design
Designed for a primary data centre and a disaster recovery site.
Local support
Comptech engineers in Bangladesh for deployment and support.
Pilot and training
A focused pilot on one use case, plus training for platform and application teams.
Explore
Other Comptech products
Comptech Container Platform
Enterprise Kubernetes with GitOps, a private registry and DC/DR.
Comptech AI Platform
Private generative AI and MLOps on your own GPUs.
Comptech Data Pipeline
Open lakehouse with SQL, pipelines, dashboards and lineage.
Comptech DocuVault
Document management for banks: capture, document AI, workflows and records.
Comptech AIOps
Monitoring, service desk, AI root cause and DR assurance for bank infrastructure.
Comptech VisionVault
On-prem AI video intelligence, transcription and evidence management.
Comptech SecOps
Security for every workload across your Comptech platforms.
Request a pilot
Tell us which applications and models you want to bring under control first. We'll plan an AI Gateway pilot around them.
Related services