SuperagentOpen-Weight Models
Run Guard on your own infrastructure with no API calls
Open-Weight Models
We publish Guard models on HuggingFace that you can run on your own infrastructure. No API calls, no data leaving your environment. 50-100ms latency on CPU or GPU.
Available Models
| Model | Parameters | Format | Use Case |
|---|---|---|---|
| superagent-guard-0.6b | 0.6B | Safetensors | Fast inference, edge deployment |
| superagent-guard-1.7b | 1.7B | Safetensors | Balanced speed and accuracy |
| superagent-guard-4b | 4B | Safetensors | Maximum accuracy |
GGUF (CPU Inference)
Quantized versions optimized for CPU inference with llama.cpp:
| Model | Parameters | Download |
|---|---|---|
| superagent-guard-0.6b-gguf | 0.6B | HuggingFace |
| superagent-guard-1.7b-gguf | 1.7B | HuggingFace |
| superagent-guard-4b-gguf | 4B | HuggingFace |
When to Self-Host
Self-host the models when you need:
- Data residency: Keep all data on your infrastructure
- Air-gapped environments: No external API calls
- Cost control: No per-request pricing at scale
- Latency guarantees: Predictable inference times
Getting Started
Download a model from HuggingFace and run it with your preferred inference framework (vLLM, llama.cpp, Transformers, etc.).
# Clone a model
git lfs install
git clone https://huggingface.co/superagent-ai/superagent-guard-1.7b
# Or download GGUF for CPU
huggingface-cli download superagent-ai/superagent-guard-1.7b-ggufBrowse all models: huggingface.co/superagent-ai