OAICA is a ternary LLM platform that runs on phones, laptops, and GPU clusters. Same API. Same model. Different hardware. $0.05/M input, $0.12/M output.
Self-hosted AI that actually runs everywhere — from your phone to your GPU cluster.
Run OAICA-700M on a phone, OAICA-35B on a cluster. Same model, same API, different hardware.
1.58-bit weights give 3-6× speedup on CPU. No GPU required for production inference.
Drop-in replacement for any OpenAI SDK. /v1/chat/completions, function calling, streaming.
Own your data. Run locally or on your infrastructure. No vendor lock-in.
Hot-swap adapters at runtime. Build custom expertise without retraining base model.
Start on our cloud API, deploy to your infrastructure when ready. Same model, same API.
Metered pricing. Pay for what you use. No subscription tax.
| Model | Input $/M | Output $/M | Context |
|---|---|---|---|
| OAICA kat-awq | $0.05 | $0.12 | 262K |
| OAICA-700M (self-hosted) | $0 | $0 | 4K |
| OAICA-3B (self-hosted) | $0 | $0 | 8K |
| OAICA-35B (self-hosted) | $0 | $0 | 32K |
Cloud API requires API key. Self-hosted is free — you only pay for hardware.
Choose the right size for your use case. All models share the same API.
~140 MB ternary deploy. 35-50 tok/s on CPU. Perfect for edge devices, IoT, mobile apps.
~1.1 GB. 20-30 tok/s on laptop. General-purpose assistant, coding, document analysis.
GPU cluster model. Front-end quality for complex reasoning, agentic workflows, RAG.
One command. No GPU required for inference.
curl -fsSL https://oaica.com/install.sh | bash
oaica run kat-awqOr self-host with Docker:
docker run -p 8080:8080 ghcr.io/sprapp/prism-engine:latestGet your API key and start building. Self-hosted option available for enterprise.