Glue-LLM: A High-Level Python SDK for Simplifying LLM Tool Calls and Multi-Agent Workflows

Glue-LLM is a high-level Python SDK designed to simplify tool calls, structured outputs, and multi-agent workflows for large language models (LLMs). It aims to help developers build complex LLM applications with less code, eliminate friction from handling low-level details, and allow developers to focus on business logic.

As vendors like OpenAI and Anthropic release models with tool-calling capabilities, developers face many challenges when integrating this capability into production-grade applications: they need to handle low-level details such as function definition serialization, model output parsing, error retries, and multi-tool call orchestration. Glue-LLM's design goal is precisely to eliminate these friction points, allowing developers to focus on business logic without worrying about the underlying plumbing.

Glue-LLM's core features include:

1. Automatic Tool Execution: Define Python functions via decorators. The SDK automatically converts function signatures to JSON Schema, constructs model calls, and parses return values—completing the request-to-result flow with one line of code.
2. Structured Output Support: Use Pydantic models to define output structures, ensuring model returns strictly adhere to type constraints and eliminating the fragility of string parsing.
3. Multi-Agent Workflow: Provide concise APIs to define agent roles, orchestrate dialogue flows, and manage state transfer—making the building of multi-agent systems intuitive and efficient.

Glue-LLM adopts a modular architecture: the core layer handles interactions with APIs of major model providers; the abstraction layer unifies tool-calling format differences across vendors; the application layer provides high-level APIs for developers, ensuring scalability. The project is built on Python, leveraging type hints and asynchronous programming features to support non-blocking parallel task execution and improve throughput in high-concurrency scenarios.

Glue-LLM is suitable for various scenarios:

• Intelligent customer service systems: Coordinate seamless integration of dialogue management, knowledge retrieval, order query, and other tools;
• Data analysis field: Structured outputs ensure SQL queries or chart configurations generated by models meet expected formats;
• Automated workflows: Multi-agent support enables complex task decomposition and parallel execution.

Compared to mature frameworks like LangChain, Glue-LLM has a more focused positioning: it does not provide peripheral functions such as vector storage or document loading, and instead focuses on the ultimate simplification of core capabilities for tool calls and multi-agent systems. For projects that do not require a complete MLOps toolchain and need to quickly integrate LLM tools, it is a lighter and more user-friendly choice.

Glue-LLM represents the evolution direction of LLM SDKs: reducing development barriers through higher-level abstraction. As model capabilities improve, efficiently transforming them into reliable applications becomes a key challenge, and Glue-LLM's solution is worth the attention and trial of LLM application developers.