tool_schema_generator: Dart LLM Tool Schema Auto-generator, Zero Boilerplate Code for OpenAI/Claude/Gemini Integration

tool_schema_generator is a code generation tool designed specifically for Dart developers, aiming to solve the tedious problem of manually writing and maintaining JSON Schema for LLM tool calls in AI Agent development. It automatically generates OpenAI, Anthropic, and Gemini-compatible tool schemas from Dart functions via annotations, supporting type inference, runtime injection, and full type safety. It is a powerful tool for building AI Agents in the Dart ecosystem.

When building AI Agents using LLMs like Gemini, OpenAI, or Claude, you need to provide the model with descriptions of callable tools. The traditional approach is to manually write and maintain JSON mappings in vendor-specific formats, which is not only tedious but also error-prone. tool_schema_generator addresses this pain point by allowing developers to write standard Dart functions with annotations to automatically generate accurate LLM tool schemas.

1. Zero Boilerplate Code: Infers type, name, and nullability from Dart syntax; the function signature serves as the single source of truth for the schema. 2. Full Type Support: Covers basic types, collections, enums, custom classes, etc. 3. Multi-Vendor Support: Generates OpenAI, Anthropic, and Gemini format schemas from the same Dart function. 4. Seamless Integration: Built on source_gen; outputs .g.dart files and coexists with json_serializable. 5. High Customization: Override name and description via @Tool(), extract information from document comments, add parameter descriptions via @Describe(), and hide application control parameters via @Inject().

Installation: Add the tool_schema_generator dependency and build_runner dev dependency to pubspec.yaml. Define Tool: Mark Dart functions with the @Tool() annotation and add a part file declaration. Generate Code: Run dart run build_runner build to generate .g.dart files. Get Schema: Use toolRegistry.schemasFor to get the corresponding vendor format. Execute Call: Use toolRegistry.call to invoke functions based on the tool name and parameters returned by the LLM.

Advanced Features: Enums are automatically converted to string enum schemas; custom classes generate nested schemas via constructor parameters; @Tool() overrides metadata; restrict support for specific vendors; @Inject() injects application control parameters (not present in the schema but available during calls). Exception Handling: Provides typed exceptions such as ToolNotFoundException, MissingToolArgumentException, InvalidToolArgumentException, and ToolExecutionException for precise error handling.

Design Philosophy: Type safety first (catch errors at compile time), single source of truth (function definition = business logic + API contract + schema template), progressive adoption (start with a single tool and coexist with existing generators). Core Value: Improve development efficiency (eliminate repetitive work), ensure type safety (consistency between code and schema), cross-platform support (three major LLM platforms), and enhance maintainability (schema automatically syncs with function changes).

Applicable Scenarios: AI assistant development (adding email sending capabilities, etc.), automated workflows, intelligent customer service systems, internal tool platforms. Summary: tool_schema_generator fills an important gap in AI Agent development for the Dart ecosystem. It has precise problem positioning, an elegant technical solution, perfect ecosystem integration, multi-platform support, and is production-ready. It is a mature LLM tool schema generation solution for building AI applications with Dart/Flutter.