Description

This issue is a designated 3-point challenge in the MemU PR Hackathon — 2026 New Year Challenge.
Participants are invited to submit pull requests between 00:00 Jan 8 and 00:00 Jan 18, 2026 (Pacific Time).
Pull requests must be submitted within this 10-day challenge period to be considered valid. Reviews will be completed within 2 days after the challenge ends, and only merged PRs will be counted toward scoring
To help us track participation and scoring, please also take 2 minutes to [register here].
A total cash prize pool of $3000 USD will be awarded based on final hackathon scores.
MemU is designed with a pluggable model architecture that supports different LLM and embedding providers.
This issue focuses on adding support for Gemini as a model provider.

Requirements

Your pull request should include:

• Complete model integration
  • Gemini LLM support
  • If the provider requires specific input/output formatting or conversion, they must be implemented as part of the integration
  • If the provider offers official embedding models, embedding support should be implemented as part of the integration
  • If the provider offers official vision models, support for vision input should be implemented as part of the integration
• Client Backend
  • You may use OpenAI SDK if the provider is fully OpenAI-compatible, but you must ensure all input data and parameters are handled correctly
• No breaking changes
  • Existing workflows must continue to work without modification
• Basic test cases
  • Each supported modality must run successfully through MemU’s full workflow (provider → memory/retrieval → inference), not just isolated unit calls.
  • If input format conversion is required, all necessary conversions (e.g. message schema, vision input structure, embedding inputs) must be correctly implemented and covered by tests.
  • Tests or runnable examples should clearly show that each modality works end-to-end without manual intervention.
• Documentation
  • How to configure and use Gemini as a model provider
  • Required environment variables or configuration options

Review Criteria

Submissions will be evaluated based on:

• Usability
  • Clear provider interface
  • Easy configuration and usage
• Stability
  • No regressions in existing functionality