Akshay66777/agentscope-chat

AgentScope Chat on Hugging Face Spaces

A single-file Gradio app running an AgentScope Agent in-process — no FastAPI server, no React frontend, just Gradio + AgentScope. That makes it eligible for HF Spaces' free zero-a10g (ZeroGPU) hardware.

Features

• —Streaming chat with an AgentScope ReActAgent
• —Built-in tools: bash, read, write, edit, glob, grep — the agent can read and modify files in its workspace
• —OpenAI-compatible: works with OpenAI or any OpenAI-compatible endpoint (set OPENAI_BASE_URL)
• —MCP / custom skills / custom OpenAI-compatible API: all supported by AgentScope's framework — extend app.py to wire them in

Setup (Hugging Face Spaces)

1. 1.Create a Space with SDK = Gradio and hardware = ZeroGPU (free)
2. 2.Add these Secrets under Settings → Variables and secrets:

1. 1.Push these files to the Space:
2. 2.app.py — the Gradio app
3. 3.requirements.txt — Python deps
4. 4.pyproject.toml — AgentScope package metadata
5. 5.src/ — the AgentScope source (installed editable)
6. 6.README.md — this file (Space frontmatter above)

1. 1.The Space auto-builds and serves the app on port 7860.

Local run

cd /home/z/my-project/agentscope
pip install -e ".[rag,tools]"
pip install -r requirements.txt
export OPENAI_API_KEY=sk-...
python app.py
# → http://localhost:7860

Extending

The app is a single app.py — edit freely:

• —Add MCP servers → pass mcps=[MCPClient(...)] to Toolkit(...) (see agentscope.mcp.StdioMCPConfig / HttpMCPConfig)
• —Add custom skills → pass skills_or_loaders=["/path/to/skill-dir"] to Toolkit(...) (skill dir must contain SKILL.md)
• —Add custom tools → subclass agentscope.tool.ToolBase and add to the tools=[...] list
• —Stream tokens → swap agent.reply() for agent.reply_stream() and yield chunks into the Gradio chatbot
• —Add long-term memory → pass middlewares=[Mem0Middleware(...)] to Agent(...)

Source

Upstream: https://github.com/agentscope-ai/agentscope