Provider key reference

Why use this reference

Pi Agent supports a growing number of providers, but adding any of them raises the same three questions:

• Where is the official sign-up page? A search for “GLM API,” for example, returns many similar-looking pages. Zhipu Open Platform and Zhipu Qingyan, its chat service, look alike but use entirely different URLs.
• Where is the API keys menu? Each dashboard organizes it differently. OpenAI uses Settings → API keys, and Anthropic uses Settings → API keys behind the gear in the lower-left corner. GLM calls the page “API Key Management,” while DeepSeek puts API keys in the left sidebar.
• What should I enter in Pi Agent? A key alone is not enough. In the Models panel, you must also choose the correct API mode—openai-completions, openai-responses, anthropic-messages, or google-generative-ai—and provide the correct baseUrl and model ID. A mistake in any field can make Test fail.

Chapter 5 creates your first OpenRouter key, and Chapter 6 adds it to Pi Agent. Chapter 10 explains when an additional provider is worthwhile, while Chapter 11 compares the available routes.

Four steps for configuring a provider

Although the seven providers have different dashboards, the setup process is broadly the same:

1. Create an account

   Registration requirements vary by provider and region. Some providers support email, Google, Apple, or GitHub sign-in; others require phone or identity verification. Check the provider’s current eligibility requirements instead of assuming that a particular country code or sign-in method is supported.

2. Set up billing or use an available free allowance

   Trial credit, free requests, rate limits, payment methods, currencies, and regional eligibility change over time. Review the provider’s current pricing and account pages. A payment method accepted by one provider may not be accepted by another.

3. Create a key under “API keys” or “Developer Center”

   The exact location differs by dashboard and is listed in each provider section below. The shared rule is important: the complete key is normally shown only when it is created. After you leave the page, the dashboard usually displays only a few identifying characters. Copy the key immediately and save it in a password manager such as Bitwarden, 1Password, or Keychain.

4. Add the key in the Pi Agent Models panel, then run Test

   Open Models in the upper-right corner of Pi Agent, select Add Provider, enter the name, API mode, base URL, and API key, add a model entry, and select Test. A green check within 3–5 seconds means the connection succeeded; a failure indicator means that at least one field needs attention. Chapter 6 walks through the process in detail.

GLM (Zhipu Qingyan)

GLM is Zhipu AI’s flagship model family, positioned as a cost-effective option with reasoning support and a 200k context window. OpenRouter is the recommended first route in this guide; see Chapter 5. Configure GLM directly only when its native service meets a specific need, and confirm current models, limits, pricing, and account eligibility in the official documentation.

Typical Add Provider settings:

Name:           GLM
API mode:       openai-completions
Base URL:       https://open.bigmodel.cn/api/paas/v4
API key:        (paste the value you just copied)
Model name:     glm-4.6
thinkingFormat: zai

OpenAI

OpenAI operates the API behind ChatGPT. The main 2026 examples listed here are the gpt-5.6 family—gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna—while the earlier gpt-4o family remains available but is no longer the first choice for new projects. Model IDs, pricing, availability, and regional billing support can change, so verify each one in OpenAI’s current documentation.

Anthropic

Anthropic develops the Claude family, which is particularly strong at programming and long-form reasoning. Its API supports native thinking blocks, which return reasoning in structured form, and prompt caching, which can reduce the cost of repeated prefixes. In this appendix, it is the only provider that does not use openai-completions for a direct connection.

Typical Add Provider settings:

Name:           Anthropic
API mode:       anthropic-messages <- Important: not openai-completions
Base URL:       (leave blank to use the official default)
API key:        sk-ant-...
Model name:     claude-sonnet-4-6

DeepSeek

DeepSeek is an AI laboratory founded by the Chinese company High-Flyer. Its API is OpenAI-compatible and is known for competitive pricing. Warning: Model catalogs change. This configuration uses the official direct IDs deepseek-chat and deepseek-reasoner; verify their status in the current documentation, especially when consulting lifecycle notices dated 2026-07-24 or later.

Groq

Groq—not xAI’s Grok—is a US company known for LPU inference hardware and speed. Some large open models are described as achieving throughput above 500 tokens per second, or 5–10 times that of some alternatives, but actual performance and free-tier availability vary. Warning: Groq’s catalog changes frequently; several Llama 3.x and older Kimi models were retired during the first half of 2026.

MiniMax

MiniMax develops models with long-context and reasoning capabilities. This edition uses MiniMax-M3, a 2025–2026 flagship example with a 1M context window and OpenAI- and Anthropic-compatible routes. The international and mainland-China services use separate endpoints.

OpenRouter

OpenRouter is an API gateway rather than a model developer. One key and one base URL provide access to models from many providers, including OpenAI, Anthropic, Google, Meta, Mistral, xAI, Perplexity, DeepSeek, GLM, and Cohere. It is the recommended first provider for this guide because it reduces initial account setup. Model availability, routing, and billing terms depend on the current OpenRouter catalog.

Provider comparison

Start with OpenRouter, then use this table for a quick scan when you need a direct connection. Confirm every model ID and account requirement in the provider’s current documentation:

Common setup problems

Most registration and connection problems fall into the following categories:

1. Test fails after you add the provider

   Check these fields in order: (1) Does the base URL have an extra trailing /? For example, .../v4/ instead of .../v4 may produce a 404. (2) Does the key contain leading or trailing spaces? Paste it into a plain-text editor to inspect it before saving. (3) Is the API mode correct? Use anthropic-messages for Anthropic and openai-completions for the other 6 routes in this appendix; the wrong protocol may return 401 or 400. (4) Is the model ID exact and current? Entering a model such as glm-4.6 incorrectly—by omitting .6 and using glm-4—or encountering a catalog change involving deepseek-chat after 2026-07 or llama-3.3-70b-versatile, can produce a 404 model-not-found error. See the troubleshooting table in Chapter 6 for more detail.

2. Your payment method is not accepted

   Handle billing by provider: (1) OpenAI and Anthropic—review the provider’s supported regions and accepted payment methods, then use an eligible method issued in your own name. (2) GLM, DeepSeek, and MiniMax—do not ask another person to fund the account or misrepresent your region; use the direct service only when its terms support you, or check whether OpenRouter legitimately offers an alternative such as meta-llama/llama-3.3-70b-instruct or deepseek/deepseek-chat. (3) MiniMax—choose the international service at platform.minimax.io only if it supports your region and payment method; the mainland-China service has separate eligibility. (4) Do not use payment workarounds of any kind, including borrowed accounts, cards, or identities. (5) OpenRouter can consolidate several providers under one eligible account, but verify its 5.5% platform fee, 5.0% cryptocurrency fee, and regional terms before paying.

3. Phone verification stalls because no message arrives

   Verification methods vary by provider and region. (1) Confirm the country code and phone number. (2) Check blocked or spam messages. (3) Wait 5 minutes before requesting another code. (4) Review whether email or federated sign-in is officially supported. Do not bypass the provider’s regional or identity requirements.

4. You cannot open the GLM, DeepSeek, or MiniMax site from your home network

   Some networks block or cannot route *.cn domains. Try three diagnostic checks: (1) Open the official site from another network to determine whether local DNS or firewall policy is responsible. (2) Review your DNS policy before trying public resolvers such as 8.8.8.8 or 1.1.1.1. (3) Use OpenRouter only when its current catalog and terms support the model and your region. Changing networks or DNS does not make an unsupported account eligible.

5. You closed the page before copying the new key

   All 7 providers follow the same security pattern: the complete key is shown only when it is created. Later, the dashboard displays only the first and last few characters. A key that was not copied cannot be recovered. Disable it, create a replacement, and save the new value in a password manager before continuing. This is a security feature, not a bug.

6. A provider passed Test but fails a few days later

   There are three common causes: (1) Your free allowance or account balance is exhausted. Check the provider dashboard and add funds only through an accepted method if needed. (2) The model has been retired. This is particularly common on Groq; consult the provider’s model catalog and use the documented successor. (3) The key was disabled after unusual activity. Review security notices in the dashboard and create a replacement key if the provider instructs you to do so.

Frequently asked questions