Build with open models
Arnict connects your application to open models through an OpenAI-compatible API. Use it to generate responses, add an assistant to your product, or power a step in an agent workflow.
Start with a model from the catalog. Each model page brings together its public API ID, capabilities, context window, availability and token rates. Choose a model marked Available, then make a request with your Arnict API key.
Keep your integration familiar
The OpenAI SDK can send chat-completion requests to Arnict. Set the base URL to https://api.arnict.com/v1, supply your Arnict key, and use the complete model ID from the catalog.
You can receive a complete response or stream text as it arrives. The quickstart includes Node.js, Python and cURL examples, installation commands, and instructions for running your first request. Optional capabilities, such as tool calling and image inputs, depend on the model you choose.
Keep control of your content
Inference prompts and responses are processed to answer your requests. They are never archived as conversation history or used for model training or advertising. Short-lived caches support request processing and expire automatically.
Arnict retains account and usage metadata needed to operate the service. Your usage dashboard shows records such as model, token counts, request status, latency and cost. It does not contain a prompt or response archive. Your application manages any conversation history it needs.
Read the data-handling guide for the distinction between temporary request content and retained account records.
Understand the cost of a request
Models have separate rates for input, cached input and output tokens, quoted per million tokens. The catalog shows current rates alongside any announced offers or standard rates.
Cached input is part of the total input count. Cached tokens use the cached input rate; the remaining input tokens use the input rate. Generated tokens use the output rate. Your dashboard records the cost for each request using confirmed usage and the applicable rates.
Use the pricing page to compare models, and the pricing guide to understand the calculation. Requests with no confirmed usage are not charged. An interrupted request may still have confirmed token usage.
Coming next: your own models
The next planned services extend Arnict beyond the shared model catalog:
- LoRA fine-tuning: adapt a base model to your task with trainable adapters and your own examples.
- Full fine-tuning (FFT): update a model’s weights using your training dataset.
- Dedicated GPU inference: run a custom model of your choice, including a model trained outside Arnict.
These features are coming soon. Configuration options, prices and setup guides will be published before access opens. Training datasets and checkpoints will have their own storage, access and deletion rules, separate from the inference-content policy.
Explore fine-tuning or dedicated inference to learn more and follow availability updates.
Make your first call
Create an account, verify your email, and generate an API key. Choose an available model, follow the quickstart, and review your first request in the dashboard.
If you need help, email support@arnict.com with the model ID, request ID and error code. Keep credentials and private prompt content out of the message.