Portkey Docs
HomeAPIIntegrationsChangelog
  • Introduction
    • What is Portkey?
    • Make Your First Request
    • Feature Overview
  • Integrations
    • LLMs
      • OpenAI
        • Structured Outputs
        • Prompt Caching
      • Anthropic
        • Prompt Caching
      • Google Gemini
      • Groq
      • Azure OpenAI
      • AWS Bedrock
      • Google Vertex AI
      • Bring Your Own LLM
      • AI21
      • Anyscale
      • Cerebras
      • Cohere
      • Fireworks
      • Deepbricks
      • Deepgram
      • Deepinfra
      • Deepseek
      • Google Palm
      • Huggingface
      • Inference.net
      • Jina AI
      • Lingyi (01.ai)
      • LocalAI
      • Mistral AI
      • Monster API
      • Moonshot
      • Nomic
      • Novita AI
      • Ollama
      • OpenRouter
      • Perplexity AI
      • Predibase
      • Reka AI
      • SambaNova
      • Segmind
      • SiliconFlow
      • Stability AI
      • Together AI
      • Voyage AI
      • Workers AI
      • ZhipuAI / ChatGLM / BigModel
      • Suggest a new integration!
    • Agents
      • Autogen
      • Control Flow
      • CrewAI
      • Langchain Agents
      • LlamaIndex
      • Phidata
      • Bring Your own Agents
    • Libraries
      • Autogen
      • DSPy
      • Instructor
      • Langchain (Python)
      • Langchain (JS/TS)
      • LlamaIndex (Python)
      • LibreChat
      • Promptfoo
      • Vercel
        • Vercel [Depricated]
  • Product
    • Observability (OpenTelemetry)
      • Logs
      • Tracing
      • Analytics
      • Feedback
      • Metadata
      • Filters
      • Logs Export
      • Budget Limits
    • AI Gateway
      • Universal API
      • Configs
      • Multimodal Capabilities
        • Image Generation
        • Function Calling
        • Vision
        • Speech-to-Text
        • Text-to-Speech
      • Cache (Simple & Semantic)
      • Fallbacks
      • Automatic Retries
      • Load Balancing
      • Conditional Routing
      • Request Timeouts
      • Canary Testing
      • Virtual Keys
        • Budget Limits
    • Prompt Library
      • Prompt Templates
      • Prompt Partials
      • Retrieve Prompts
      • Advanced Prompting with JSON Mode
    • Guardrails
      • List of Guardrail Checks
        • Patronus AI
        • Aporia
        • Pillar
        • Bring Your Own Guardrails
      • Creating Raw Guardrails (in JSON)
    • Autonomous Fine-tuning
    • Enterprise Offering
      • Org Management
        • Organizations
        • Workspaces
        • User Roles & Permissions
        • API Keys (AuthN and AuthZ)
      • Access Control Management
      • Budget Limits
      • Security @ Portkey
      • Logs Export
      • Private Cloud Deployments
        • Architecture
        • AWS
        • GCP
        • Azure
        • Cloudflare Workers
        • F5 App Stack
      • Components
        • Log Store
          • MongoDB
    • Open Source
    • Portkey Pro & Enterprise Plans
  • API Reference
    • Introduction
    • Authentication
    • OpenAPI Specification
    • Headers
    • Response Schema
    • Gateway Config Object
    • SDK
  • Provider Endpoints
    • Supported Providers
    • Chat
    • Embeddings
    • Images
      • Create Image
      • Create Image Edit
      • Create Image Variation
    • Audio
      • Create Speech
      • Create Transcription
      • Create Translation
    • Fine-tuning
      • Create Fine-tuning Job
      • List Fine-tuning Jobs
      • Retrieve Fine-tuning Job
      • List Fine-tuning Events
      • List Fine-tuning Checkpoints
      • Cancel Fine-tuning
    • Batch
      • Create Batch
      • List Batch
      • Retrieve Batch
      • Cancel Batch
    • Files
      • Upload File
      • List Files
      • Retrieve File
      • Retrieve File Content
      • Delete File
    • Moderations
    • Assistants API
      • Assistants
        • Create Assistant
        • List Assistants
        • Retrieve Assistant
        • Modify Assistant
        • Delete Assistant
      • Threads
        • Create Thread
        • Retrieve Thread
        • Modify Thread
        • Delete Thread
      • Messages
        • Create Message
        • List Messages
        • Retrieve Message
        • Modify Message
        • Delete Message
      • Runs
        • Create Run
        • Create Thread and Run
        • List Runs
        • Retrieve Run
        • Modify Run
        • Submit Tool Outputs to Run
        • Cancel Run
      • Run Steps
        • List Run Steps
        • Retrieve Run Steps
    • Completions
    • Gateway for Other API Endpoints
  • Portkey Endpoints
    • Configs
      • Create Config
      • List Configs
      • Retrieve Config
      • Update Config
    • Feedback
      • Create Feedback
      • Update Feedback
    • Guardrails
    • Logs
      • Insert a Log
      • Log Exports [BETA]
        • Retrieve a Log Export
        • Update a Log Export
        • List Log Exports
        • Create a Log Export
        • Start a Log Export
        • Cancel a Log Export
        • Download a Log Export
    • Prompts
      • Prompt Completion
      • Render
    • Virtual Keys
      • Create Virtual Key
      • List Virtual Keys
      • Retrieve Virtual Key
      • Update Virtual Key
      • Delete Virtual Key
    • Analytics
      • Graphs - Time Series Data
        • Get Requests Data
        • Get Cost Data
        • Get Latency Data
        • Get Tokens Data
        • Get Users Data
        • Get Requests per User
        • Get Errors Data
        • Get Error Rate Data
        • Get Status Code Data
        • Get Unique Status Code Data
        • Get Rescued Requests Data
        • Get Cache Hit Rate Data
        • Get Cache Hit Latency Data
        • Get Feedback Data
        • Get Feedback Score Distribution Data
        • Get Weighted Feeback Data
        • Get Feedback Per AI Models
      • Summary
        • Get All Cache Data
      • Groups - Paginated Data
        • Get User Grouped Data
        • Get Model Grouped Data
        • Get Metadata Grouped Data
    • API Keys [BETA]
      • Update API Key
      • Create API Key
      • Delete an API Key
      • Retrieve an API Key
      • List API Keys
    • Admin
      • Users
        • Retrieve a User
        • Retrieve All Users
        • Update a User
        • Remove a User
      • User Invites
        • Invite a User
        • Retrieve an Invite
        • Retrieve All User Invites
        • Delete a User Invite
      • Workspaces
        • Create Workspace
        • Retrieve All Workspaces
        • Retrieve a Workspace
        • Update Workspace
        • Delete a Workspace
      • Workspace Members
        • Add a Workspace Member
        • Retrieve All Workspace Members
        • Retrieve a Workspace Member
        • Update Workspace Member
        • Remove Workspace Member
  • Guides
    • Getting Started
      • A/B Test Prompts and Models
      • Tackling Rate Limiting
      • Function Calling
      • Image Generation
      • Getting started with AI Gateway
      • Llama 3 on Groq
      • Return Repeat Requests from Cache
      • Trigger Automatic Retries on LLM Failures
      • 101 on Portkey's Gateway Configs
    • Integrations
      • Llama 3 on Portkey + Together AI
      • Introduction to GPT-4o
      • Anyscale
      • Mistral
      • Vercel AI
      • Deepinfra
      • Groq
      • Langchain
      • Mixtral 8x22b
      • Segmind
    • Use Cases
      • Few-Shot Prompting
      • Enforcing JSON Schema with Anyscale & Together
      • Detecting Emotions with GPT-4o
      • Build an article suggestion app with Supabase pgvector, and Portkey
      • Setting up resilient Load balancers with failure-mitigating Fallbacks
      • Run Portkey on Prompts from Langchain Hub
      • Smart Fallback with Model-Optimized Prompts
      • How to use OpenAI SDK with Portkey Prompt Templates
      • Setup OpenAI -> Azure OpenAI Fallback
      • Fallback from SDXL to Dall-e-3
      • Comparing Top10 LMSYS Models with Portkey
      • Build a chatbot using Portkey's Prompt Templates
  • Support
    • Contact Us
    • Developer Forum
    • Common Errors & Resolutions
    • December '23 Migration
    • Changelog
Powered by GitBook
On this page
  • How Custom Webhook Works
  • Headers
  • Webhook Response

Was this helpful?

Edit on GitHub
  1. Product
  2. Guardrails
  3. List of Guardrail Checks

Bring Your Own Guardrails

PreviousPillarNextCreating Raw Guardrails (in JSON)

Last updated 9 months ago

Was this helpful?

Portkey supports bringing your own Guardrail using a custom webhook:

You can add Webhook as a Guardrail Check and setup any Guardrail Actions along with it. This is useful when you have an existing custom guardrail pipeline in your app and you are sending your LLM inputs or outputs to it for sync or async evaluation. Now, you can bring that onto Portkey's Gateway, make it production-grade, and enforce LLM behavior in real-time, easily.

How Custom Webhook Works

In the Guardrail check, Portkey expects your Webhook URL and any headers (including Authorization) you need to send.

Headers

Headers here should be a JSON object, like below:

{
    "Authorization":"MY_API_KEY",
    "Content-Type":"Application/JSON"
}

Portkey makes a POST request to your webhook URL and expects two objects in the response: verdict and data.

Webhook Response

Portkey expects two objects: verdict and data

Object
Type
Required?

verdict

boolean

Yes

data

any

No

Here's a sample webhook response:

{
  "verdict": true,
  "data": {
    "reason": "User passed all security checks",
    "score": 0.95,
    "additionalInfo": {
      "userStatus": "verified",
      "lastCheckTimestamp": "2024-08-21T14:30:00Z"
    }
  }
}

Check out the Webhook implementation here:

Based on the verdict value, the Guardrail Check will PASS or FAIL, and will have subsequent impact on the Guardrail Actions you've set.

The webhook request automatically time out after 3 seconds - this can not be changed. So, if a webhook request times out, the Guardrail verdict will return PASS for that request.

Head over to the if you are building out custom webhooks and need any help!

Portkey Discord community
https://github.com/Portkey-AI/gateway/blob/main/plugins/default/webhook.ts
import {
  HookEventType,
  PluginContext,
  PluginHandler,
  PluginParameters,
} from '../types';
import { post } from '../utils';

function parseHeaders(headers: unknown): Record<string, string> {
  try {
    if (typeof headers === 'object' && headers !== null) {
      return headers as Record<string, string>;
    }
    if (typeof headers === 'string') {
      try {
        const parsed = JSON.parse(headers as string);
        return parsed;
      } catch {
        throw new Error('Invalid headers format');
      }
    }
    return {};
  } catch (error: any) {
    throw error;
  }
}

export const handler: PluginHandler = async (
  context: PluginContext,
  parameters: PluginParameters,
  eventType: HookEventType
) => {
  let error = null;
  let verdict = false;
  let data: any = null;
  const transformedData: Record<string, any> = {
    request: {
      json: null,
      text: null,
    },
    response: {
      json: null,
      text: null,
    },
  };
  let transformed = false;

  try {
    const url = parameters.webhookURL;

    if (!url) {
      throw new Error('Missing webhook URL');
    }

    // Validate URL format
    try {
      new URL(url);
    } catch {
      throw new Error('Invalid webhook URL format');
    }

    let headers: Record<string, string>;
    try {
      headers = parseHeaders(parameters.headers);
    } catch (e: any) {
      throw new Error(`Failed to parse headers: ${e.message}`);
    }

    const requestBody = {
      ...context,
      // Setting headers to undefined to avoid passing sensitive information to the webhook endpoint.
      // This can later be controlled through parameters.
      request: { ...context.request, headers: undefined },
      eventType,
    };

    const response = await post(
      url,
      requestBody,
      { headers },
      parameters.timeout || 3000
    );
    verdict = response.verdict;

    if (
      response.transformedData?.request?.json &&
      eventType === 'beforeRequestHook'
    ) {
      transformedData.request.json = response.transformedData.request.json;
      transformed = true;
    }

    if (
      response.transformedData?.response?.json &&
      eventType === 'afterRequestHook'
    ) {
      transformedData.response.json = response.transformedData.response.json;
      transformed = true;
    }

    data = {
      verdict,
      explanation: verdict
        ? 'Webhook request succeeded'
        : 'Webhook request failed',
      webhookUrl: url,
      responseData: response.data,
      requestContext: {
        headers,
        timeout: 3000,
      },
    };
  } catch (e: any) {
    error = e;
    delete error.stack;

    data = {
      explanation: `Webhook error: ${e.message}`,
      webhookUrl: parameters.webhookURL || 'No URL provided',
      requestContext: {
        headers: parameters.headers || {},
        timeout: 3000,
      },
    };
  }

  return { error, verdict, data, transformedData, transformed };
};