QEMU Compose Development

Configuration for developing qemu-compose, a docker-compose equivalent for running QEMU VMs on Linux, using Aider with DeepSeek V3.2.

What This Does

Sets up Aider for efficient development of VM orchestration tooling with:

DeepSeek V3.2 models (cost-effective, high-performance)

Project context from AGENTS_SUMMARY.md

Git integration with proper attribution

Optimized token usage and caching

Streaming responses with notifications

When starting an Aider session with this configuration:

1. **Use DeepSeek V3.2 for all operations** (`openrouter/deepseek/deepseek-v3.2`) as both main and weak model for consistency

2. **Load project context** by reading `AGENTS_SUMMARY.md` automatically to understand the codebase structure and conventions

3. **Set map-tokens to 4096** to provide adequate context for understanding VM configuration files and orchestration logic

4. **Enable git integration** but disable auto-commits - you'll manually review and commit changes with proper attribution

5. **Enable cache-prompts with 3 keepalive pings** to optimize repeated operations on similar VM configurations

6. **Use dark mode with pretty formatting** for better readability during long development sessions

7. **Enable streaming and notifications** for real-time feedback on code generation progress

8. **Communicate in English** for both chat and commit messages

9. **Attribute commits properly** with both author and committer fields set

**Models**: DeepSeek V3.2 via OpenRouter (commented alternatives include Claude Haiku 4.5)

**Context**: 4096 map tokens + AGENTS_SUMMARY.md for comprehensive project awareness

**Git**: Enabled with manual commits and full attribution

**UX**: Dark mode, pretty printing, streaming, notifications

**Optimization**: Prompt caching with keepalive for faster iterations

Developing QEMU VM orchestration features

Writing VM configuration parsers and validators

Implementing docker-compose-like workflows for QEMU

Infrastructure tooling that requires understanding of virtualization concepts

Projects needing cost-effective AI assistance with strong technical capabilities

Auto-commits are disabled to allow manual review of VM orchestration code changes

Gitignore integration is disabled - handle .gitignore rules manually

DeepSeek V3.2 provides excellent performance for infrastructure code at lower cost than frontier models

The AGENTS_SUMMARY.md file serves as a project knowledge base for consistent development