Lakera

Lakera is an AI security company building runtime defenses for generative AI applications. Its flagship Lakera Guard API screens prompts and responses for prompt injection, jailbreaks, PII leakage, unsafe content, and policy violations, while Lakera Red provides automated red teaming and risk assessment for GenAI systems. Lakera follows an API-first architecture with a managed SaaS platform, regional endpoints, self-hosted deployments for regulated environments, and a free Gandalf training game that has driven much of the prompt injection research community.

3 APIs 8 Features

AI SecurityLLM SecurityPrompt InjectionGuardrailsRed TeamingGenAIAPI

APIs

Lakera Guard

Lakera Guard is a low-latency screening API that inspects text content sent to or from LLMs and flags threats including prompt injection, jailbreaks, PII, profanity, and policy ...

Lakera Red

Lakera Red is an automated red teaming product that probes GenAI applications for jailbreaks, prompt injection, data leakage, and other adversarial failures, then produces a ris...

Gandalf

Gandalf is Lakera's free interactive game that teaches prompt injection by challenging players to extract a secret from an LLM across progressively harder defenses. It has becom...

Collections

Lakera Guard API

OPEN

Pricing Plans

Lakera Plans Pricing

1 plans

PLANS

Rate Limits

Lakera Rate Limits

2 limits

RATE LIMITS

FinOps

Lakera Finops

FINOPS

Features

Prompt Injection Detection

Detects direct and indirect prompt injection attempts targeting LLM-powered applications and agents.

Jailbreak Detection

Identifies attempts to bypass system prompts, safety policies, and model guardrails.

PII Detection

Screens prompts and responses for personally identifiable information leakage.

Content Moderation

Flags unsafe, toxic, or policy-violating content in user inputs and model outputs.

Custom Policies

Configurable detection policies scoped to projects, models, and use cases.

Regional Endpoints

Hosted endpoints in US, EU, and APAC regions for data residency and latency.

Self-Hosted Deployment

On-premises and VPC deployments for regulated industries that cannot send data to a SaaS.

MCP Server

Model Context Protocol server for embedding Lakera Guard into AI clients like Claude Code and Cursor.

Use Cases

LLM Application Guardrails

Screen inputs and outputs of chatbots, copilots, and RAG applications for prompt injection and unsafe content.

AI Agent Protection

Inspect tool inputs, retrieved context, and agent reasoning steps for indirect prompt injection.

Workforce AI Security

Govern employee use of public LLMs and prevent sensitive data exfiltration.

GenAI Red Teaming

Run automated adversarial assessments against pre-production GenAI applications.

Regulated Industry Deployments

Self-host Guard in financial services, healthcare, and government environments with strict data residency.

Integrations

OpenAI

Wrap OpenAI calls with Guard screening for input and output safety.

Anthropic

Screen prompts and responses from Anthropic Claude models.

LangChain

Drop-in callbacks and chains for integrating Guard into LangChain applications.

LlamaIndex

Screen queries and retrieved context in LlamaIndex RAG pipelines.

Model Context Protocol

MCP server for surfacing Lakera Guard inside Claude Code, Cursor, and other MCP clients.

Kubernetes

Self-hosted Helm-based deployments with health, readiness, and liveness probes.

Resources

Documentation

APIReference

GettingStarted

GitHubOrganization

GitHubOrganization

Sources

opencollection: 1.0.0
info:
  name: Lakera Guard API
  version: '2'
request:
  auth:
    type: bearer
    token: '{{bearerToken}}'
items:
- info:
    name: Guard
    type: folder
  items:
  - info:
      name: Screen text content for threats
      type: http
    http:
      method: POST
      url: https://api.lakera.ai/v2/guard
      body:
        type: json
        data: '{}'
    docs: Submit messages to be screened. Lakera Guard returns a flagged decision when prompt injection, jailbreak, PII, unsafe
      content, or other policy violations are detected.
  - info:
      name: Retrieve detailed detector results
      type: http
    http:
      method: POST
      url: https://api.lakera.ai/v2/guard/results
      body:
        type: json
        data: '{}'
    docs: Returns the per-detector breakdown that informed a Guard decision, useful for debugging policies and tuning thresholds.
- info:
    name: Policies
    type: folder
  items:
  - info:
      name: Validate policy configuration (self-hosted)
      type: http
    http:
      method: GET
      url: https://api.lakera.ai/v2/policies/health
    docs: Validate policy configuration (self-hosted)
  - info:
      name: Lint a policy file (self-hosted)
      type: http
    http:
      method: POST
      url: https://api.lakera.ai/v2/policies/lint
      body:
        type: json
        data: '{}'
    docs: Lint a policy file (self-hosted)
- info:
    name: Health
    type: folder
  items:
  - info:
      name: Kubernetes startup probe (self-hosted)
      type: http
    http:
      method: GET
      url: https://api.lakera.ai/v2/startupz
    docs: Kubernetes startup probe (self-hosted)
  - info:
      name: Kubernetes readiness probe (self-hosted)
      type: http
    http:
      method: GET
      url: https://api.lakera.ai/v2/readyz
    docs: Kubernetes readiness probe (self-hosted)
  - info:
      name: Kubernetes liveness probe (self-hosted)
      type: http
    http:
      method: GET
      url: https://api.lakera.ai/v2/livez
    docs: Kubernetes liveness probe (self-hosted)
bundled: true