Why Your Base URL Matters
The base URL is the most common point of failure when switching providers. Most OpenAI SDKs default to https://api.openai.com. If you change the API key but forget to update the base URL, your requests go to the wrong server. This results in authentication errors or unexpected model availability. Always check the provider's documentation for the exact endpoint. For example, a drop-in uncensored API might use https://api.apirouteropen.com/v1. Changing just the base URL and key allows you to keep your existing code. This ensures compatibility with libraries like LangChain or LlamaIndex. Verify the URL before scaling up your usage.
The Model ID Misconception
Developers often assume the model ID corresponds to a vendor's name like gpt-4 or claude-3. In reality, the model ID is a specific routing key defined by the gateway. For our uncensored service, the ID is simply uncensored. Sending gpt-3.5-turbo might fail or return a different model if the gateway doesn't map it. Always use the exact ID provided in the documentation. This applies to all LLM proxies. Check the GET /v1/models endpoint to see available options. Misidentifying the model ID causes immediate request failures. It is not a vendor's product but a configuration parameter.
Rate Limit Surprises
Rate limits are often per API key, not per account. If you use a single key for multiple processes, you might hit the limit quickly. Our API allows 300 requests per minute and 8 concurrent requests per key. Exceeding these limits returns a 429 error. This is critical for high-volume applications. Use separate keys for different services to avoid interference. Monitor your usage through the dashboard. Rate limits protect server stability but can disrupt automated workflows. Plan your concurrency carefully. A single key replacement invalidates the old key immediately. This prevents concurrent usage of the same key.
Streaming vs. Non-Streaming Confusion
Streaming responses require specific client handling. The SDK must be configured to handle server-sent events (SSE). If you use a non-streaming client, you will wait for the entire response. This increases latency for long outputs. Our API supports streaming with token usage in the last chunk. This allows real-time display of text. Non-streaming is better for batch processing. Choose the mode based on your use case. Incorrect handling can lead to hanging connections. Always test streaming with a small sample first. Ensure your client supports SSE for optimal performance.
Token Counting Errors
Token counting varies between providers. Some count whitespace, others do not. Our API counts prompt and completion tokens together. The context window is 100,000 tokens. Max output is 16,000 tokens per request. If you exceed this, the request fails. Do not assume OpenAI's tokenization matches ours. Use the provider's tokenizer for accuracy. Token counting errors lead to unexpected charges or failures. Always verify token counts before sending large prompts. This is especially important for long-context applications. Miscounting can waste credits on rejected requests.
Ignoring Function Calling Limits
Function calling requires specific JSON formatting. The SDK must support the response_format parameter. Our API supports JSON mode for structured outputs. If you ignore this, the model may return invalid JSON. This breaks automated parsing. Always test function calling with a simple schema first. Ensure your client handles tool choices correctly. Function calling is not supported in all models. Check the documentation for supported features. Ignoring these limits leads to unreliable automation. Verify the output format matches your expectations.
Crypto Top-up Delays
Crypto payments can take time to confirm. We accept USDT (TRC20) and USDC (Base). Payments from $20 to $500 are supported. Bonus credit applies at $50, $100, and $200. Delays occur due to blockchain congestion. Check the transaction hash for status. Our system credits your account after confirmation. This usually takes a few minutes. During high traffic, it may take longer. Use a reliable blockchain explorer. Do not resend the same transaction. This causes duplicate credits or errors. Plan your top-ups in advance for uninterrupted service.
Trial Credit Expiration
Trial credits expire after 7 days. New accounts receive $0.50. This allows testing without commitment. However, the short window limits extensive testing. Use the trial to verify compatibility. Check base URL, model ID, and streaming. If it works, top up with crypto. Trial credits are non-refundable. They are ideal for quick validation. Do not rely on them for long-term projects. The expiration is strict. Plan your testing schedule accordingly. This ensures you don't waste time during the trial period.
