Using Your Own API Key

Beaver allows you to use your own API key for unlimited access. This bypasses Beaver's built-in credits system and bills usage directly to your provider account.

Why is login required with my own API key? All requests are processed through Beaver's server. Your computer connects to our server, which then calls the AI provider. This enables many features and is central to how Beaver works. Your API key is stored locally, transmitted securely, and never saved on our servers.

Supported Providers

Beaver supports API keys from three major AI providers:

Adding Your API Key

Why Beaver needs more from your API key

A key that works fine for direct chat requests can still fail in Beaver. Two things make Beaver much more demanding than a typical chat request:

1. Beaver works with a lot of context. As an AI agent built for academic researchers, Beaver reads and processes full or partial papers alongside your question, so a single request can send tens of thousands of tokens.
2. Beaver is an AI agent that makes multiple requests per question. Answering one question often involves 5-10 back-and-forth calls.

Together, this means Beaver uses 10x, 100x, or even 1000x more tokens than a simple request. A key that works fine for short chats can be rejected by the provider when used with Beaver because the provider enforces rate, volume, or tier limits on your account.

If your key struggles with this load, your options are to upgrade your provider tier, switch to a provider with more generous limits (OpenAI tends to be the most forgiving but requires verification), or use Beaver credits.

Common Issues

Google/Gemini: Rate limits, model overloaded, or slow responses

Google/Gemini API keys frequently hit rate limits, report the model as overloaded, or time out. These errors come directly from Google and affect all applications, not just Beaver. Because of Beaver's higher token demands, Gemini keys are especially prone to these errors.

1. Free API keys are extremely limited and not appropriate for Beaver. Upgrade to a paid key.

2. Paid Tier 1 keys can still hit limits when Google servers are overloaded or under heavy use. Retry after a moment, or upgrade to Tier 2 or higher.

3. Try OpenAI instead. OpenAI is more generous with rate limits and often works better than a Tier 1 Gemini key.

Anthropic: Rate limits or insufficient credits

This error means your Anthropic API key has either run out of credits or exceeded its rate limit.

1. Check your balance in the Claude Console and add credits if needed.

2. Rate limits on Tier 1 keys cap input tokens at 30,000 per minute (ITPM). Because Beaver is an AI agent that makes repeated requests, this limit can be exhausted quickly. Upgrade to Tier 2 or higher, switch to Haiku 4.5 (which has higher rate limits), or use a different provider.

OpenAI: Verification required

OpenAI requires identity verification before allowing streaming (real-time responses):

"OpenAI requires verification to use this model with streaming."

Custom Models and Endpoints

Using custom models and endpoints is an advanced feature and should only be used by experienced users. Many models or endpoints are incompatible with Beaver and will not work correctly. Some of these issues show up as error messages in Beaver, and others may lead to degraded performance or incorrect responses. While we always try to return useful error messages, it is impossible to test every model and endpoint. As a result, error messages may be misleading, incomplete and even wrong.