llm_tool_firebase_ai 0.4.0

Use llm_tool with Firebase AI Logic (firebase_ai). Turns your @LlmTool functions into Gemini function declarations and runs the whole tool loop.

0.4.0

• Sends @Param limits from llm_tool 0.8.0: minimum, maximum, minItems and maxItems as firebase_ai schema fields. firebase_ai has no fields for minLength, maxLength and pattern, so they are added to the parameter's description, e.g. "Airport code (exactly 3 characters, matching ^[A-Z]{3}$)". call() and invoke() check them all.
• The re-export of llm_tool also leaves out the new LlmToolset annotation.
• Requires llm_tool 0.8.0.

0.3.0

• Renamed from llm_tool_calling_firebase_ai (now discontinued), to go with the core package's new name, llm_tool. Import package:llm_tool_firebase_ai/llm_tool_firebase_ai.dart.
• toFirebaseAITool() replaces toFirebaseAiTool() (Dart style for two-letter acronyms); the old name is deprecated until 1.0.0. The extensions are now ToolDefinitionFirebaseAI and ToolListFirebaseAI.
• The re-export of llm_tool_calling leaves out the annotations (LlmTool, Param and the deprecated Tool).
• Requires llm_tool_calling 0.7.0.

0.2.0

• Uses ToolDefinition.invoke from llm_tool_calling 0.6.0, so tools behave the same with every SDK. ToolConfirmation now comes from llm_tool_calling (still available through this package).
• If a tool throws during firebase_ai's automatic function calling, Gemini now gets a clear error.
• Requires llm_tool_calling 0.6.0.

0.1.1

• Works with llm_tool_calling 0.5.x (and still 0.3.x and 0.4.x).
• Clearer package description: the adapter runs the whole tool loop.

0.1.0

• Initial release.
• chat.sendMessageWithTools(message, allTools) runs the whole tool loop. Tested live with gemini-3.8-flash.
• toolResponses(results) sends tool results with the role user, because newer Gemini models reject the function role that firebase_ai's Content.functionResponses uses.
• allTools.toFirebaseAiTool() for firebase_ai's own automatic function calling (fails with newer models until firebase_ai releases flutterfire#18685; see the README).
• toFunctionDeclarations() and respondTo(functionCall) for manual function calling.
• Arguments are validated before anything runs; errors go back to Gemini.
• Tools with requiresConfirmation only run when a confirm callback approves them.
• Tools with duplicate names are refused with an ArgumentError, instead of firebase_ai silently keeping only one.
• Re-exports ToolDefinition, ToolArgumentException and the other runtime types, so a model-setup file only needs firebase_ai and this package.
• toFirebaseJsonSchema() converts llm_tool_calling JSON schemas to firebase_ai JSONSchema, sent as parametersJsonSchema.