Big integer precision loss in tool calling

[cuihaoleo]

Describe the bug

ADK seems to have a precision limitation for integer arguments in function calling, which may catch Python users in surprise.

To Reproduce

A simplistic example:

from google.adk.agents.llm_agent import Agent

def add(a: int, b: int) -> int:
    """Returns the sum of two integers."""
    return a + b

root_agent = Agent(
    model='gemini-2.5-flash',
    name='root_agent',
    description="Add two integers.",
    instruction="""
    You are a helpful assistant that adds two integers.
    Call the `add` tool to perform the calculation.
    Just return its result. Do not doubt its correctness.
    """,
    tools=[add],
)

Interaction:

Me: 10000000000000001+123456789
Agent: The sum of 10000000000000001 and 123456789 is 10000000123456788.

Function calling log:

{"parts":[{"function_call":{"args":{"a":1e16,"b":123456789},"name":"add"},"thought_signature":"......"}],"role":"model"}
{"parts":[{"function_response":{"name":"add","response":{"result":1.0000000123456788e16}}}],"role":"user"}

You can see the numbers are actually represented as float numbers behind the sense, causing a loss of precision and inaccurate result.

Expected behavior

The integers are passed to tools without precision loss, and 10000000000000001+123456789 = 10000000123456790.

I believe this is caused by JSON serialization, which does not distinguish int and float numbers and may support only limited precision for numbers. But it's quite surprising for Python (and maybe even Go/Java) users, as the number I used (10000000000000001) is well within the range of 64-bit int types.

Desktop (please complete the following information):

• OS: Linux
• Python version(python -V): 3.13.7
• ADK version(pip show google-adk): 1.18.0

Model Information:

• Are you using LiteLLM: No
• Which model is being used: gemini-2.5-flash but the issue is model-agnostic I think.

[cuihaoleo]

To put things in perspective, openai-agents library appears to keep integers precise:

import asyncio

from agents import Agent, Runner, function_tool

@function_tool
def add(a: int, b: int) -> int:
    """Returns the sum of two integers."""
    print(f"Adding {a} and {b}")
    return a + b

agent = Agent(
    name="Addition Agent",
    instructions="""
    You are a helpful assistant that adds two integers.
    Call the `add` tool to perform the calculation.
    Just return its result. Do not doubt its correctness.
    """,
    tools=[add],
)

async def main():
    result = await Runner.run(agent, input="10000000000000001+123456789")
    print(result.final_output)


if __name__ == "__main__":
    asyncio.run(main())

Adding 10000000000000001 and 123456789
10000000123456790

[yyyu-google]

I see the bug is in function call part from backend API

{"function_call":{"args":{"a":1e16,"b":123456789},"name":"add"

"a" should be 10000000000000001 not 1e16.

I tested with curl command to locate the root cause

Step 1: put the following in request.json file

{
  "contents": [
    {
      "parts": [
        {
          "text": "what is the result of 10000000000000001 + 123456789?"
        }
      ],
      "role": "user"
    }
  ],
  "tools": [
    {
      "functionDeclarations": [
        {
          "description": "Add two integers.",
          "name": "add_integers",
          "parameters": {
            "properties": {
              "a": {
                "type": "INTEGER"
              },
              "b": {
                "type": "INTEGER"
              }
            },
            "required": [
              "a",
              "b"
            ],
            "type": "OBJECT"
          },
          "response": {
            "type": "INTEGER"
          }
        }
      ]
    }
  ],
  "generationConfig": {}
}

Step 2: do the following curl command

curl -X POST \
     -H "Authorization: Bearer $(gcloud auth print-access-token)" \
     -H "Content-Type: application/json" \
     "https://us-central1-aiplatform.googleapis.com/v1/projects/my-project/locations/us-central1/publishers/google/models/gemini-2.5-flash:generateContent" \
     -d @request.json

Step 3: raw output from backend

{
  "candidates": [
    {
      "content": {
        "role": "model",
        "parts": [
          {
            "functionCall": {
              "name": "add_integers",
              "args": {
                "b": 123456789,
                "a": 1e+16
              }
            },
            "thoughtSignature": "CtEBAePx/168/WpvhWLPDkVOibAVTdsHbeeyE82iomH+yr4nXndM9FZaZINPa1G0TRO5r+iuwG7NwPpA2JCYF3/RdCCW9juckiFEhxxz88MwdG08gSXJ8HSLe16s1yalyTPPMpZrenY8BUc8jBNRrTfNXVsGd6YZ0oTYevTXBh1bAsh5h8a5SgqV2C2pjXsqbTFYKiG2H3ox9oPWSLK2nQX3ACQssoTuVW6YlTai09CJuHnoP9C0IBln6knC5jtfr+nLUFtrsU2+h0Bg72uPjoJq0D4="
          }
        ]
      },
      "finishReason": "STOP",
      "score": -7.2109456062316895,
      "avgLogprobs": -1.0301350866045271
    }
  ],
  "usageMetadata": {
    "promptTokenCount": 50,
    "candidatesTokenCount": 7,
    "totalTokenCount": 98,
    "billablePromptUsage": {
      "textCount": 76
    },
    "trafficType": "ON_DEMAND",
    "promptTokensDetails": [
      {
        "modality": "TEXT",
        "tokenCount": 50
      }
    ],
    "candidatesTokensDetails": [
      {
        "modality": "TEXT",
        "tokenCount": 7
      }
    ],
    "thoughtsTokenCount": 41
  },
  "modelVersion": "gemini-2.5-flash",
  "createTime": "2025-11-19T21:56:16.430337Z",
  "responseId": "AD0eaYGiGtHKnvgPvezU8Ac"
}

The arg "a" is 1e+16 not 10000000000000001