Using Downsizing with Zapier AI Agents

Prerequisites

A Zapier account (any plan that allows Webhooks/API Request steps) and a Downsizing account that can make requests (pay-per-use or a connected subscription both work). Grab your Router ID and API key from the Inference API page.

Why there's no one-click setup

Zapier's current AI surfaces — AI by Zapier (the built-in action used inside Zaps and Zapier Agents), Zapier Chatbots' "use your own API key" option, and Zapier MCP— don't expose a custom base-URL or OpenAI-compatible-endpoint field:

• AI by Zapier's "Bring Your Own Key"tier only lets you pick a fixed provider — OpenAI, Anthropic, Google Gemini, Azure OpenAI, or Amazon Bedrock — and connect an account for it. There's a provider dropdown and a model dropdown, but no field to override the provider's endpoint.
• Zapier Chatbots' own-API-key option is the same shape: paste an OpenAI or Anthropic key, pick a model — no base URL override.
• Zapier MCPruns in the opposite direction from what you might expect: it lets an external AI assistant (Claude, ChatGPT, etc.) call Zapier's 6,000+ app actions as tools. It's not a way to swap the model that powers Zapier's own AI features, so it doesn't help you route Zapier's AI calls through Downsizing either.

AI Actions, Zapier's older ChatGPT-plugin-era integration, is no longer being developed — Zapier's own docs point people to Zapier MCP instead, and it doesn't change this picture since it never had a custom-endpoint field.

In short: there is no first-class "add your own LLM provider" slot in Zapier today. The reliable way to route a Zap or Agent's model calls through Downsizing is a generic HTTP action that calls Downsizing's API directly.

Manual Setup

Add a Webhooks by Zapier (or API Request) action as a step in your Zap, or as a custom action inside a Zapier Agent, and configure it as a POST request:

1. Add a step, search for Webhooks by Zapier (or the built-in API Request (Beta) action), and choose POST.
2. Set the URL to your Downsizing Anthropic Messages endpoint:

https://api.downsizing.dev/YOUR_ROUTER_ID/anthropic/v1/messages

Replace YOUR_ROUTER_ID with your workspace UUID from the dashboard.

3. Add these headers:

Content-Type: application/json
x-api-key: YOUR_DOWNSIZING_API_KEY
anthropic-version: 2023-06-01

Replace YOUR_DOWNSIZING_API_KEY with your key (starts with dws_) from the same Inference API page.

4. Set the request body to JSON, following Anthropic's Messages API shape — wire in data from earlier steps (e.g. a trigger field or a previous action's output) wherever you'd normally insert user input:

{
  "model": "claude-opus-4-8",
  "max_tokens": 1024,
  "messages": [
    { "role": "user", "content": "{{input}}" }
  ]
}

Set Model Nameto whichever Claude or GPT model you want Downsizing to route — Downsizing selects the actual upstream model itself, so any recognizable name works. Parse the reply's content[0].textfield in your next step to use the model's response elsewhere in the Zap.

Verify

Use Zapier's built-in Test step on the Webhooks/API Request action and confirm you get a 200 response with a completion back. Then check your savings dashboard to see the request appear in your usage metrics.