Book a demo
Book a demo
EnglishРусский中文

Platform

Data, agents and models under shared policies

Every Prizma decision is recorded in one history: which request arrived, which policy applied and what the model received.

Book a demo
  • Shared policies
  • One decision history
  • Your infrastructure

Data protection

Inspect text and attachments in requests and model responses.

  • Personal data and secrets
  • Masking and blocking
  • Documents and images
Request with an attachmentClaude

Supply_agreement.pdftext recognised

  • Counterparty name[NAME_1]
  • Bank account[ACCOUNT_1]
  • Portal passwordblocked

The task reached the model without the original values

Model routing

Connect providers through a shared API and choose the destination for each request.

  • Cloud and local models
  • OpenAI and Anthropic API compatibility
  • Fallback routes on provider failure
Engineering team routeOpenAI API · Anthropic API
  • 1Cloud modelunavailable
  • 2Local modelfallback active

The request moved to an approved fallback route

Agent actions

Approve changes before a connected tool is called.

  • Allow, request approval or deny
  • Approval for the exact action
  • Task-scoped, time-limited access
MCP tool rulesClaude Code
  • tracker.read_issuereadallowed
  • tracker.update_issuechangepending approval
  • repo.delete_branchdeletedenied

Permissions are granted until the task is complete

Knowledge bases

Check documents before publication and control the available version.

  • Admission and quarantine
  • Versions and publication approval
  • Links to source material
Knowledge base versionsPolicies
  1. v12PublishedUsed by applicationsactive
  2. v13In reviewTwo documents in quarantinequarantine

The new version becomes available after approval

In-house protection models

Decisions follow your company’s policies and take signals from Prizma’s in-house protection models into account.

  • Company policies
  • Protection model signals
Prizma decision

Decision history

Investigate events and send them to your security monitoring system.

  • Request, policy and result
  • Audit integrity checks
  • Event export to SIEM
Audit event10:42:18
Source
Cursor · engineering team
Policy
Personal data → mask
Result
Sent with replacements

Audit integrity verifiedSent to SIEM

More about the platform

More capabilities

Capabilities that extend the main areas and are configured for your environment.

Targeted stop controls

Stop selected requests while keeping permitted routes available.

  • User, model or agent controls
  • Disable cloud destinations
  • Operator action history

Proprietary code

Recognise fragments of company algorithms before they reach a model.

  • Protected asset registration
  • Snippet matching
  • Transfer blocking and event history

MCP tools

Manage tool servers and inspect data at the point of invocation.

  • Server and permission registry
  • Argument and result inspection
  • Approval for write operations

Agent memory

Keep record provenance and inspect memory when stored and recalled.

  • Source and trust level
  • Quarantine for suspicious records
  • Memory integrity verification

Budgets and usage

Attribute spend to request owners and control consumption.

  • Team and user budgets
  • Token and cost accounting
  • Request rate limits

Enterprise identity

Connect Prizma to your organisation’s identity management.

  • SAML and OIDC sign-in
  • SCIM users and groups
  • Roles, API keys and MFA

AI security assessment

Examine models, system instructions and agent permissions.

  • Leakage and prompt injection tests
  • Agent-to-resource consequence graph
  • AI model and component inventory

Operations

Deploy and monitor Prizma in your company infrastructure.

  • Containers and Kubernetes
  • Monitoring and alerting
  • Redundancy, updates and recovery

Prompt injection

Detect attempts to override model roles, bypass policies or extract system instructions.

  • Direct and indirect attacks
  • Instructions embedded in tool results
  • Protection model signals combined with company policies

Documents and images

Inspect attachment text and information recognised in images.

  • English and Russian OCR
  • Isolated document processing
  • Image region redaction

Streaming responses

Inspect responses as they stream, including text and tool call arguments.

  • Streaming fragment inspection
  • Text channel controls
  • Unsafe transfer interruption

Need something that is not on the list?

Tell us about your case

Start with a pilot

Pick one workflow. We’ll walk through the data to protect, what agents may do and how Prizma connects to your systems.

Discuss a pilot sales@hackadvisor.io