Gemini
Google Gemini is used for specific tasks in the Green Goods ecosystem where its strengths complement the primary Claude-based toolchain. This page documents current integration points and usage patterns.
Current Use Cases
Long-Context Document Analysis
Gemini's large context window makes it suitable for analyzing lengthy documents that exceed Claude's context limits:
- Grant application review -- Full grant proposals with appendices can be processed in a single pass
- Specification comparison -- Comparing multiple large specification documents side-by-side
- Transcript analysis -- Processing lengthy meeting transcripts when the full context is needed at once
Multimodal Tasks
Gemini's vision capabilities are used for:
- Screenshot analysis -- Evaluating UI screenshots against design specifications
- Document OCR -- Extracting structured data from scanned documents or images
- Visual regression -- Comparing before/after screenshots of component changes
Search-Grounded Research
Gemini's integration with Google Search provides grounded research capabilities for:
- Ecosystem research -- Understanding the current state of protocols Green Goods integrates with (EAS, Hats Protocol, Hypercerts)
- Dependency evaluation -- Researching library updates, breaking changes, and migration paths
- Standards tracking -- Monitoring EIP/ERC proposals relevant to the protocol
When to Use Gemini vs Claude
| Task | Recommended | Reason |
|---|---|---|
| Code implementation | Claude | Deeper reasoning, TDD workflow integration |
| Code review | Claude | Better judgment, lower false-positive rate |
| Long document analysis | Gemini | Larger context window |
| Web-grounded research | Gemini | Search integration |
| Screenshot evaluation | Either | Both have strong vision |
| Architecture decisions | Claude | Access to full .claude/ context |
| Mechanical transforms | Either | Both handle well |
Integration Approach
Gemini is used as a standalone tool rather than integrated into the agent pipeline. Common access patterns:
- Google AI Studio -- For interactive exploration and prompt testing
- API calls -- For automated tasks in scripts or workflows
- Vertex AI -- For production workloads with SLA requirements
The project does not maintain Gemini-specific configuration files (unlike .claude/ for Claude or .codex/ for Codex). Gemini tasks use the project's CLAUDE.md and AGENTS.md as context input where applicable.
Limitations in This Project
- No persistent memory or session continuity (unlike Claude Code's agent memory)
- No direct integration with the
.claude/skills/system - Cannot execute tools against the local codebase (use Claude Code for that)
- No path-scoped rule loading -- context must be manually provided
Cost Considerations
Gemini is generally more cost-effective for high-volume, lower-complexity tasks. For the Green Goods project:
- Use Gemini for research and analysis where grounding matters
- Use Claude for implementation and review where codebase context matters
- Avoid using Gemini for tasks that require the full
.claude/context stack
Next best action
Understand how Model Context Protocol connects AI tools to the codebase.
MCP Guide