README.md

metadata

title: One-Shot Deploy
emoji: 🚀
colorFrom: indigo
colorTo: purple
sdk: docker
pinned: true

One-Shot Deploy

Describe a demo in plain English → Crusoe's multi-model agent writes the code and ships it to Hugging Face Spaces in minutes.

How it works

Stage	Model	What happens
1 · Intent	DeepSeek R1	Reasons about your prompt, selects a template, writes the AI system prompt
2 · Codegen	Kimi-K2	Writes a complete single-file Streamlit app wired to Crusoe inference
3 · Validate	(ast parser)	Checks syntax and structural patterns
4 · Deploy	Qwen3 (healer)	Pushes to HF Spaces, injects secrets, auto-fixes on build failure (up to 3×)

Required secrets

Set these in Settings → Variables and secrets in your Space:

Name	Description
CRUSOE_API_KEY	Crusoe Managed Inference API key
HF_TOKEN	Hugging Face token with write scope
HF_USERNAME	Your Hugging Face username

Optionally override the default models:

Name	Default
CRUSOE_BASE_URL	https://api.crusoe.ai/v1
INTENT_MODEL	deepseek-ai/DeepSeek-R1
CODEGEN_MODEL	moonshotai/Kimi-K2-Instruct
HEALER_MODEL	Qwen/Qwen3-235B-A22B

Built by

Crusoe DevRel · Powered by Crusoe Managed Inference