Adding an Integration

This page is a preview

The AI dispatch / integration layer hasn't been built yet. This page describes the planned architecture and how you'll be able to add new integrations.

The Plan

autosolve will have an executor layer that launches external AI tools when rules match. An integration is a concrete executor — ra-aid, Ollama, a custom script, etc.

What an Integration Will Look Like

Based on the current architecture patterns, an integration will likely:

1. Implement an Executor interface
2. Receive issue/PR context and produce a result
3. Be registered in DI like workers and preflights

// Hypothetical — not implemented yet
type Executor interface {
    Execute(ctx context.Context, task Task) (Result, error)
}

How You Can Help Now

The integration layer is in the design phase. If you have ideas or want to help shape it:

• Open an issue with your use case — what AI tool, what triggers, what output format
• Check the Ideas & Wishlist — the Rule Engine and Multi-Agent sections are relevant
• Read the Idea — the original vision describes the executor concept

Current Extension Points

While the integration layer is being designed, you can already extend autosolve by:

• Adding a Worker — new polling or processing tasks
• Adding infrastructure clients in internal/infrastructure/ — new API clients, storage backends
• Adding new partition types — see Schedule Package

Conventions

All code follows the project's Review Guidelines. Key points for infrastructure code:

• Concrete names — no Service, Manager, Handler
• Rate limiting via HTTP transport (transparent to callers)
• Error interfaces for classification (Retryable, ServicePressure, WaitHinted)
• Tests required for pkg/ and infrastructure code