OctoTools
NormalDescription
OctoTools is an agentic framework with extensible tools for complex reasoning, featuring a tool card system for flexible composition of diverse reasoning capabilities.
Key Features
- Training-free agentic framework — no fine-tuning needed; works out of the box with any supported LLM
- Standardized tool cards — encapsulate tool functionality with metadata for plug-and-play integration of new tools
- Hierarchical planner — handles both high-level strategy and low-level step-by-step action refinement
- Task-specific toolset optimization — learns the best subset of tools for downstream tasks automatically
- Broad LLM support — compatible with OpenAI, Anthropic, Gemini, DeepSeek, Grok, vLLM, Ollama, and LiteLLM
- Proven accuracy gains — achieves 9.3% average accuracy improvement over GPT-4o across 16 diverse tasks
Use Cases
💡 Complex multi-step reasoning — solving math, science, and medical questions that require tool-augmented reasoning
💡 Benchmark evaluation — testing and comparing LLM performance across diverse reasoning tasks
💡 Research prototyping — quickly assembling tool-augmented agents for academic experiments
💡 Domain-specific problem solving — combining calculators, search engines, code executors, and custom tools for specialized tasks
Categories
Quick Start
pip install octotoolkit → Import Planner and Executor from octotools → Initialize with your LLM engine (e.g. GPT-4o) → Define tool cards for your task → Run planner.solve(task) to execute multi-step reasoning