Hugging Face

Scira uses the Hugging Face Inference Providers Router at router.huggingface.co/v1 — an OpenAI-compatible endpoint that routes requests to 15+ third-party inference providers (DeepInfra, Groq, Fireworks, Together, Nebius, SambaNova, and more). Your HF access token is the only credential you need; the router handles provider selection.

Get a token

Create a Hugging Face account

Go to huggingface.co and sign up. A free account works.

Create an access token

Open Settings → Access Tokens → Create new token. Select a token with at least Read permissions. Copy the token — it starts with hf_.

Add it to Scira

scira init
# or manually:
echo "HF_API_KEY=hf_..." >> ~/.scira/.env

Configure

~/.scira/config.json

{
  "llmProvider": "huggingface",
  "model": "meta-llama/Llama-3.3-70B-Instruct"
}

Switch to HuggingFace from the TUI with /llm → HuggingFace.

Available models

The /model picker fetches the live list from router.huggingface.co/v1/models and filters to models where at least one provider reports supports_tools: true. Fallback static list:

Model ID	Notes
`meta-llama/Llama-3.3-70B-Instruct`	Default. Strong general instruction model with tool use
`meta-llama/Llama-3.1-70B-Instruct`	Previous Llama 3.1 70B
`Qwen/Qwen2.5-72B-Instruct`	Alibaba's 72B — strong on code and reasoning
`mistralai/Mistral-7B-Instruct-v0.3`	Compact 7B, low cost
`deepseek-ai/DeepSeek-V3`	DeepSeek V3 via the router

Use the exact owner/model-name format as it appears on the Hugging Face model page. The router automatically selects the fastest available inference provider for the model; you can force a specific one by appending :provider (e.g. meta-llama/Llama-3.3-70B-Instruct:groq).

How the router picks a provider

By default the router selects the fastest available backend. Append :cheapest to a model ID to prefer the lowest-cost provider instead, or configure your preferred provider order in your HF account settings and use :preferred.

Pricing

Inference via the router is billed by the underlying provider at their standard rates. Costs vary by model and provider. Check huggingface.co/inference/models for current provider availability and pricing per model.