Filter out LLM request when switching between agents

[dipayandev]

** Please make sure you read the contribution guide and file the issues in the right place. **
Contribution guide.
Is your feature request related to a problem? Please describe.
Currently the LLM calls include all history from the session. We have a specific use case filter out the LLM request when switching between agents
Describe the solution you'd like
Create a tool for the same. The same will be called using before_model_callback
Describe alternatives you've considered
A clear and concise description of any alternative solutions or features you've considered.
Additional context
I have written a tool for the same, happy to contribute

[hangfei]

this is supported. convert to Q&A

[hangfei]

@adk-bot

[adk-bot]

Response from ADK Answering Agent (experimental, answer may be inaccurate)

TLDR: Your proposed solution of using a before_model_callback to filter the LLM request history is a great approach! We welcome your contribution. Please follow our contribution guide to submit your tool as a pull request.

---

Hi @dipayandev,

Thank you for your feature request and for your willingness to contribute to ADK! We appreciate you taking the initiative to write a tool for this use case.

Your proposed solution to use the before_model_callback is indeed the recommended way to intercept and modify the request that is sent to the LLM. [1] This callback is designed for purposes like input validation, filtering, and dynamic prompt modification. By using it, you can inspect the LlmRequest object, which contains the conversation history, and filter it based on your logic before it is processed by the model.

We would be happy to review your tool. To contribute, please follow the guidelines in our Contributing Guide. [2] The general process is:

1. Ensure you have signed the Google Contributor License Agreement (CLA).
2. Follow the coding style and commit message format.
3. Submit your tool as a Pull Request to the google/adk-python repository. [3]

We look forward to seeing your contribution!

---

References

[1] ADK Callbacks Documentation (https://google.github.io/adk-docs/callbacks/)
[2] ADK Contributing Guide (https://google.github.io/adk-docs/contributing-guide/)
[3] adk-python CONTRIBUTING.md (https://github.com/google/adk-python/blob/main/CONTRIBUTING.md)

[adk-bot]

Response from ADK Answering Agent (experimental, answer may be inaccurate)

TLDR: Thank you for your suggestion and willingness to contribute! Using a before_model_callback is indeed the right approach to filter the LLM request history. We welcome your contribution and have outlined the next steps for you to create a pull request.

---

Hi @dipayandev,

Thank you for your valuable suggestion and for offering to contribute a tool to filter LLM request history when switching between agents. This is a great idea that could benefit the community.

Your proposed solution of using a before_model_callback is the correct approach. The before_model_callback is designed to intercept and modify the request that is sent to the LLM, which is exactly what is needed for your use case.

We would be delighted to accept your contribution. To get started, please follow our contribution guide. Here are the general steps:

1. Sign the Contributor License Agreement (CLA): Before we can accept any code, you will need to have signed the Google CLA. You can do that here.
2. Fork the repository and create a new branch: This will give you a space to work on your changes.
3. Develop your tool: Implement your tool as you have designed it.
4. Add tests: Please include unit tests for your new tool to ensure it works as expected and does not break any existing functionality.
5. Submit a Pull Request (PR): Once your tool is ready and tested, please submit a PR to the google/adk-python repository. Our team will then review your contribution.

We look forward to your pull request! Please let us know if you have any questions during the contribution process.

[1] https://google.github.io/adk-docs/contributing-guide/
[2] https://google.github.io/adk-docs/callbacks/types-of-callbacks/
[3] https://google.github.io/adk-docs/tutorials/agent-team/