Overview
Bring your own key means you paste an API key from an AI provider you already pay for, and Taskade uses that key to run AI steps in your automations. The provider bills you directly for those runs, and your Taskade AI credits stay untouched.
TL;DR: Connect your own OpenAI, Anthropic Claude, Google Gemini, Groq, or OpenRouter key to an automation and every run bills to your provider account, using zero Taskade AI credits. One key field per provider, stored encrypted, reusable across every step in the workspace. See Pricing for what credits cover.
Note: The provider connectors on this page are rolling out and may not appear in your step list yet. The built-in Taskade AI steps do the same work today using your workspace AI credits.
Connect a provider
Every provider connects the same way.
- Open Automations in your workspace and click Create.
- Add a trigger, then add a step and pick your provider.
- Click Connect.
- Paste your API Key into the single key field.
- Save. Taskade checks the key against the provider before it saves, so a typo or a revoked key fails right there.
Your key is masked in the interface and stored encrypted. One connection covers the whole workspace, so every automation step that uses that provider reuses it.
Two OpenAI steps, one name. You will see two OpenAI entries in the step list. The built-in Taskade AI one runs on your workspace credits with nothing to set up. The one on this page runs on your own key and bills your OpenAI account. The logos differ, and the description on the your-key version says it bills to your account.
Where to get each key
| Provider | Get your key at | Key starts with |
|---|---|---|
| OpenAI | platform.openai.com/api-keys | sk- |
| Anthropic Claude | console.anthropic.com/settings/keys | sk-ant- |
| Google Gemini | aistudio.google.com/apikey | AIza |
| Groq | console.groq.com/keys | gsk_ |
| OpenRouter | openrouter.ai/settings/keys | sk-or- |
Sign in to the provider, create a key, copy it once, and paste it into Taskade. Most providers show a key only at the moment you create it.
What each provider can do
| Provider | Steps available |
|---|---|
| OpenAI | Ask ChatGPT, Vision Prompt, Extract Structured Data, Generate Image |
| Anthropic Claude | Ask Claude, Vision Prompt, Extract Structured Data |
| Google Gemini | Ask Gemini, Vision Prompt, Extract Structured Data |
| Groq | Ask AI, Extract Structured Data |
| OpenRouter | Ask AI, Extract Structured Data |
One OpenRouter key reaches models from many providers at once, which makes it the shortest path if you would rather manage a single account than five.
Ask the model
The Ask step sends a prompt and returns the reply as text. Type the model name yourself, so you are never limited to a list Taskade curates. Whatever your key can reach, you can use.
Vision Prompt
Vision steps take an image plus a question about it, and return a text answer. The Gemini version fetches the image itself, so give it a public https:// address. An address on a private network is refused.
Extract Structured Data
Give the step a JSON schema and a piece of text, and it returns fields you can map into later steps.
| Field | What it holds |
|---|---|
data |
The extracted values |
raw |
The model's untouched reply |
parsed |
true only when the reply matches your schema, with required fields present and types correct |
parsed is stricter than "the model returned valid JSON", so it is safe to branch on. A schema the model cannot work with fails the step with a message instead of quietly returning something unusable.
Generate Image
Available on OpenAI. The step returns a hosted url for the image, which later steps reference like any other link. Pick an image model that returns a link rather than raw image data.
Credits and billing
| Step type | Who pays | Counts against Taskade credits |
|---|---|---|
| Built-in Taskade AI steps | Taskade, from your plan | Yes |
| Your own provider key | Your provider account | No |
Your provider sets the price of each run, and you watch usage in the provider's own dashboard. Taskade never adds a markup and never spends credits on a step that runs on your key. Everything else in the automation, such as triggers, filters, and workspace actions, behaves exactly as it always does.
Common patterns
| You want to | Build this |
|---|---|
| Run high-volume AI without spending credits | Swap the Taskade AI step for the same step on your own key |
| Use a model Taskade does not offer | Connect OpenRouter and type the model name |
| Pull fields out of an email or form | Extract Structured Data, then branch on parsed |
| Describe an uploaded screenshot | Vision Prompt with a public image address |
| Compare two providers on the same prompt | Two Ask steps side by side, then a step that reads both |
Frequently Asked Questions
Do these steps use my Taskade AI credits?
No. Runs on your own key bill to your provider account and consume zero Taskade AI credits. Only the built-in Taskade AI steps draw from your workspace credit pool.
Where is my API key stored?
Encrypted, at the workspace level, and masked in the interface after you save it. Taskade uses it only to run the steps in your automations. Revoke it in your provider's dashboard at any time to cut access off immediately.
Why are there two OpenAI steps?
One runs on Taskade credits with nothing to set up. The other runs on your own OpenAI key and bills your OpenAI account. Check the description on the step: the your-key version says it bills to your account.
Which models can I use?
Any model your key can reach. You type the model name into the step rather than picking from a fixed list, so a model your provider released yesterday works today.
When is parsed true on Extract Structured Data?
Only when the reply matches the schema you supplied, with required fields present and types correct. Valid JSON alone is not enough, which is what makes parsed safe to build a branch on.
Can my AI agents use my own key too?
Yes, separately. The connectors on this page are automation steps. A custom AI agent has its own key field in the agent's settings, under AI Model, and it accepts an OpenAI or Anthropic key. That field is available on Enterprise and on AppSumo and PitchGround lifetime plans. Set the agent to a specific model rather than Auto, or it keeps running on Taskade credits. Your agents keep their built-in toolkit either way. See the Billing FAQ for the full details.
Related guides
- Automations. Triggers, steps, filters, and run history.
- AI Models and Credit Costs. What Taskade credits cover and how they are spent.
- MCP Client Integration. Call tools on a remote MCP server from an automation.
- Custom AI Agents. Build agents with their own instructions, knowledge, and tools.
- Integrations. The full catalog of triggers and actions.