gorilla: what it is, what problem it solves & why it's gaining traction

What it solves

Gorilla addresses the challenge of Large Language Models (LLMs) hallucinating when attempting to call APIs. It enables LLMs to accurately invoke thousands of APIs by ensuring the generated calls are semantically and syntactically correct, reducing errors in tool use.

How it works

Gorilla uses a combination of specialized fine-tuning and retrieval-augmented training to connect LLMs with a massive collection of APIs. The project provides a suite of tools including:

• OpenFunctions: A drop-in alternative for function calling that supports multiple languages (Python, Java, JavaScript) and REST APIs.
• GoEx (Execution Engine): A sandboxed runtime that executes LLM-generated actions with safety guarantees, featuring "post-facto validation" and undo capabilities to mitigate risk.
• API Zoo: A community-maintained repository of structured API documentation used to keep models up-to-date and reduce hallucinations.
• RAFT: A fine-tuning recipe for domain-specific Retrieval-Augmented Generation (RAG) that trains models to quote documents directly.

Who it’s for

Developers building AI agents, software engineers integrating LLMs with external services, and researchers focusing on function calling and tool-use benchmarks.

Highlights

• Massive API Support: Capable of invoking 1,600+ APIs accurately.
• Berkeley Function Calling Leaderboard (BFCL): A comprehensive benchmark for evaluating single-turn, multi-turn, and multi-step function calling.
• Agent Arena: A head-to-head comparison platform for LLM agents using an ELO rating system.
• Safe Execution: Docker-based sandboxed environment for executing API calls and filesystem operations via GoEx.
• Commercial Readiness: Offers Apache 2.0 licensed models for commercial use.