Model Catalog

Explore and query every AI model available to your workspace, with instant code snippets for all supported providers.

The Model Catalog is the evolution of Virtual Keys, providing a centralized and powerful way to manage, discover, and use AI models within your workspace. It consists of two main sections: AI Providers, where you manage your connections, and Models, where you explore what you can use.

How it Works: Inheritance from the Organization

The most significant change with the Model Catalog is the concept of inheritance. Think of it this way:

1. Your Organization Admin creates a master Integration at the company level (e.g., for “Azure Production”). They add the credentials and can set default budgets, rate limits, and an allow-list of approved models for that integration.
2. When they provision this integration to your workspace, a corresponding AI Provider is automatically created in your Model Catalog.
3. This new AI Provider in your workspace inherits all the settings from the organization-level integration, including its credentials, model access, and spending limits.

This “create once, provision many” approach provides central governance while giving workspaces the flexibility they need.

---

The Model Catalog Experience by Role

Your experience with the Model Catalog will differ based on your role within the Portkey organization.

For Workspace Members (Developers): Discover and Build

As a developer, your experience is simplified and streamlined. You primarily interact with the Models tab, which acts as your personal “Model Garden.”

• Discover Models: The “Models” tab is a complete gallery of every single model you have been given access to by your admins.
• Get Code Snippets: Click on any model, and Portkey will generate the exact code snippet you need to start making calls, with the correct provider and model slugs already included.
• Simplified API Calls: You can now call any model directly using the model parameter, formatted as @{provider_slug}/{model_slug}. This lets you switch between providers and models on the fly with a single Portkey API key.

# Switch between a model on OpenAI and one on Bedrock seamlessly
client.chat.completions.create(
    model="@openai-prod/gpt-4o", 
    messages=[...]
)

client.chat.completions.create(
    model="@bedrock-us/claude-3-sonnet-v1",
    messages=[...]
)

For Workspace Admins: Manage and Customize

As a Workspace Admin, you have more control over the providers within your workspace via the AI Providers tab.

You will see a list of providers that have been inherited from the organization. From here, you have two primary options when you click Create Provider:

1. Inherit from an Org Integration: You can create additional providers that are based on an existing org-level integration. This is useful for subdividing access within your team. For example, if your workspace has a $1000 budget on the main "Azure Prod" integration, you could create a new provider from it named "azure-prod-experimental" and give it a stricter$100 budget for a specific project.
2. Create a New Workspace-Exclusive Integration: If your Org Admin has enabled the permission, you can create a brand new integration from scratch. This provider is exclusive to your workspace and functions just like the old Virtual Keys did.

For Organization Admins: A View into Workspaces

While Org Admins primarily work in the main Integrations dashboard, the Model Catalog provides a crucial feedback loop:

When a Workspace Admin creates a new, workspace-exclusive integration (option #2 above), you gain full visibility. This new integration will automatically appear on your main Integrations page under the “Workspace-Created” tab, ensuring you always have a complete audit trail of all provider credentials being used across the organization.

SDK Integration and Advanced Usage

While the new model parameter is the recommended approach for its simplicity, Portkey maintains full backward compatibility and offers flexible integration options for various SDKs.

Remember, the term “Virtual Key” is now synonymous with the AI Provider slug found in your Model Catalog.

Using with Portkey SDK

You can set a default AI Provider at initialization or override it per request.

NodeJS

import Portkey from 'portkey-ai';

// Set a default AI Provider for the client
const portkey = new Portkey({
    apiKey: process.env.PORTKEY_API_KEY,
    provider:"@YOUR_AI_PROVIDER_SLUG"
//  provider: "@YOUR_AI_PROVIDER_SLUG"
});

// Or, override it for a specific call
const chatCompletion = await portkey.chat.completions.create({
    messages: [{ role: 'user', content: 'Say this is a test' }],
    model: 'gpt-3.5-turbo',
}, { provider:"@ANOTHER_AI_PROVIDER_SLUG" });

import Portkey from 'portkey-ai';

// Set a default AI Provider for the client
const portkey = new Portkey({
    apiKey: process.env.PORTKEY_API_KEY,
    provider:"@YOUR_AI_PROVIDER_SLUG"
//  provider: "@YOUR_AI_PROVIDER_SLUG"
});

// Or, override it for a specific call
const chatCompletion = await portkey.chat.completions.create({
    messages: [{ role: 'user', content: 'Say this is a test' }],
    model: 'gpt-3.5-turbo',
}, { provider:"@ANOTHER_AI_PROVIDER_SLUG" });

from portkey_ai import Portkey

# Set a default AI Provider for the client
portkey = Portkey(
    api_key="PORTKEY_API_KEY",
    provider="@YOUR_AI_PROVIDER_SLUG"
#   provider="@YOUR_AI_PROVIDER_SLUG"
)

# Or, override it for a specific call
completion = portkey.with_options(
    provider="@ANOTHER_AI_PROVIDER_SLUG"
).chat.completions.create(
    messages = [{ "role": 'user', "content": 'Say this is a test' }],
    model = 'gpt-3.5-turbo'
)

Using with OpenAI SDK

Simply point the OpenAI client to Portkey’s gateway and pass your AI Provider slug in the headers.

Python

from openai import OpenAI
from portkey_ai import PORTKEY_GATEWAY_URL, createHeaders

client = OpenAI(
  api_key="", # can be left blank
  base_url=PORTKEY_GATEWAY_URL,
  default_headers=createHeaders(
    api_key="PORTKEY_API_KEY",
    provider="@YOUR_AI_PROVIDER_SLUG"
#   provider="@YOUR_AI_PROVIDER_SLUG"
  )
)

from openai import OpenAI
from portkey_ai import PORTKEY_GATEWAY_URL, createHeaders

client = OpenAI(
  api_key="", # can be left blank
  base_url=PORTKEY_GATEWAY_URL,
  default_headers=createHeaders(
    api_key="PORTKEY_API_KEY",
    provider="@YOUR_AI_PROVIDER_SLUG"
#   provider="@YOUR_AI_PROVIDER_SLUG"
  )
)

import OpenAI from "openai";
import { PORTKEY_GATEWAY_URL, createHeaders } from 'portkey-ai';

const openai = new OpenAI({
  apiKey: '', // can be left blank
  baseURL: PORTKEY_GATEWAY_URL,
  defaultHeaders: createHeaders({
    apiKey: "PORTKEY_API_KEY",
    provider:"@YOUR_AI_PROVIDER_SLUG"
//  provider: "@YOUR_AI_PROVIDER_SLUG"
  })
});

Assistant

Responses are generated using AI and may contain mistakes.