Back to Apps

Agent Distillation
by nardien
Distill large LLM agents into efficient small language models with retrieval and code tools.
0 stars
Works in:claude
Exposes:Tools
What it does
Agent Distillation is a specialized framework designed to transfer the agentic capabilities of large language models (teachers) into much smaller, more efficient models (students). It focuses on mimicking complex behaviors, specifically those involving retrieval-augmented generation and code execution, allowing small models to maintain high performance on factual and mathematical reasoning tasks.
Tools
- Trajectory Logger: Captures detailed agent run logs to create high-quality training datasets.
- SFT Trainer: Uses the TRL library to fine-tune small models while maintaining compatibility with the smolagents framework.
- Benchmarking Suite: Evaluates distilled agents on standard factual and mathematical reasoning benchmarks.
Installation
{
"mcpServers": {
"agent-distillation": {
"command": "python3",
"args": ["-m", "agent_distillation.server"]
}
}
}
Supported hosts
- claude
Quick install
pip install -e .[distill]Information
- Pricing
- free
- Published
- 6/14/2026
- stars
- 0
Categories
Choose your AI client and follow the steps below.
Claude Desktop
Add the server configuration to your claude_desktop_config.json using the python3 module command.





