Spaces:
Running on CPU Upgrade
Running on CPU Upgrade
File size: 9,666 Bytes
914a44a a1a2253 914a44a 0127ea2 914a44a 7b4e2da 914a44a 7b4e2da 914a44a | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 | # HF Agent
An MLE agent CLI with MCP (Model Context Protocol) integration and built-in tool support.
## Quick Start
### Installation
```bash
# Clone the repository
git clone git@github.com:huggingface/hf_agent.git
cd hf-agent
```
#### Install recommended dependencies
```bash
uv sync --extra agent # or uv sync --extra all
```
### Interactive CLI
```bash
uv run python -m agent.main
```
This starts an interactive chat session with the agent. Type your messages and the agent will respond, using tools as needed.
The agent will automatically discover and register all tools from configured MCP servers.
## Architecture
### Component Overview
```
┌─────────────────────────────────────────────────────────────┐
│ User/CLI │
└────────────┬─────────────────────────────────────┬───────────┘
│ User request │ Events
↓ ↑
submission_queue event_queue
│ │
↓ │
┌────────────────────────────────────────────────────┐ │
│ submission_loop (agent_loop.py) │ │
│ ┌──────────────────────────────────────────────┐ │ │
│ │ 1. Receive Operation from queue │ │ │
│ │ 2. Route to Handler (run_agent/compact/...) │ │ │
│ └──────────────────────────────────────────────┘ │ │
│ ↓ │ │
│ ┌──────────────────────────────────────────────┐ │ │
│ │ Handlers.run_agent() │ ├─────────┤
│ │ │ │ Emit │
│ │ ┌────────────────────────────────────────┐ │ │ Events │
│ │ │ Agentic Loop (max 10 iterations) │ │ │ │
│ │ │ │ │ │ │
│ │ │ ┌──────────────────────────────────┐ │ │ │ │
│ │ │ │ Session │ │ │ │ │
│ │ │ │ ┌────────────────────────────┐ │ │ │ │ │
│ │ │ │ │ ContextManager │ │ │ │ │ │
│ │ │ │ │ • Message history │ │ │ │ │ │
│ │ │ │ │ (litellm.Message[]) │ │ │ │ │ │
│ │ │ │ └────────────────────────────┘ │ │ │ │ │
│ │ │ │ │ │ │ │ │
│ │ │ │ ┌────────────────────────────┐ │ │ │ │ │
│ │ │ │ │ ToolRouter │ │ │ │ │ │
│ │ │ │ │ ├─ bash │ │ │ │ │ │
│ │ │ │ │ ├─ read_file │ │ │ │ │ │
│ │ │ │ │ ├─ write_file │ │ │ │ │ │
│ │ │ │ │ └─ McpConnectionManager │ │ │ │ │ │
│ │ │ │ │ ├─ mcp__server1__* │ │ │ │ │ │
│ │ │ │ │ └─ mcp__server2__* │ │ │ │ │ │
│ │ │ │ └────────────────────────────┘ │ │ │ │ │
│ │ │ └──────────────────────────────────┘ │ │ │ │
│ │ │ │ │ │ │
│ │ │ Loop: │ │ │ │
│ │ │ 1. LLM call (litellm.acompletion) │ │ │ │
│ │ │ ↓ │ │ │ │
│ │ │ 2. Parse tool_calls[] │ │ │ │
│ │ │ ↓ │ │ │ │
│ │ │ 3. Execute via ToolRouter │ │ │ │
│ │ │ ↓ │ │ │ │
│ │ │ 4. Add results to ContextManager │ │ │ │
│ │ │ ↓ │ │ │ │
│ │ │ 5. Repeat if tool_calls exist │ │ │ │
│ │ └────────────────────────────────────────┘ │ │ │
│ └──────────────────────────────────────────────┘ │ │
└────────────────────────────────────────────────────┴─────────┘
```
### Agentic Loop Flow
```
User Message
↓
[Add to ContextManager]
↓
╔═══════════════════════════════════════╗
║ Iteration Loop (max 10) ║
║ ║
║ Get messages + tool specs ║
║ ↓ ║
║ litellm.acompletion() ║
║ ↓ ║
║ Has tool_calls? ──No──> Done ║
║ │ ║
║ Yes ║
║ ↓ ║
║ Add assistant msg (with tool_calls) ║
║ ↓ ║
║ For each tool_call: ║
║ • ToolRouter.execute_tool() ║
║ • Add result to ContextManager ║
║ ↓ ║
║ Continue loop ─────────────────┐ ║
║ ↑ │ ║
╚═════════╧═══════════════════════╧═════╝
```
## Project Structure
```
agent/
├── config.py # Configuration models
├── main.py # Interactive CLI entry point
├── context_manager/
│ └── manager.py # Message history management
└── core/
├── agent_loop.py # Main agent loop and handlers
├── session.py # Session management
├── mcp_client.py # MCP SDK integration
└── tools.py # ToolRouter and built-in tools
test_integration.py # Basic integration tests
test_tools.py # Tool execution tests
eval/ # Evaluation suite (see eval/README.md)
```
## Events
The agent emits the following events via `event_queue`:
- `processing` - Starting to process user input
- `assistant_message` - LLM response text
- `tool_call` - Tool being called with arguments
- `tool_output` - Tool execution result
- `turn_complete` - Agent finished processing
- `error` - Error occurred during processing
- `interrupted` - Agent was interrupted
- `compacted` - Context was compacted
- `undo_complete` - Undo operation completed
- `shutdown` - Agent shutting down
## Development
### Adding Built-in Tools
Edit `agent/core/tools.py`:
```python
def create_builtin_tools() -> list[ToolSpec]:
return [
ToolSpec(
name="your_tool",
description="What your tool does",
parameters={
"type": "object",
"properties": {
"param": {"type": "string", "description": "Parameter description"}
},
"required": ["param"]
},
handler=your_async_handler
),
# ... existing tools
]
```
### Adding MCP Servers
Add to your config:
```python
config = Config(
model_name="anthropic/claude-sonnet-4-5-20250929",
mcp_servers=[
MCPServerConfig(
name="your_server",
command="command",
args=["arg1", "arg2"],
env={"KEY": "value"}
)
]
)
```
|