Go microkernel platform weaving AI chat, RAG, and agents via plugins

What it does Weave is a Go application platform built on Gin that organizes itself as a microkernel wrapped around a layered core. It offers a plugin system where features—AI or otherwise—register themselves, resolve dependencies, and can be hot-plugged at runtime. The project bundles a full AI development stack including LLM routing, chat services, agent frameworks, and RAG backed by RedisSearch, all exposed through a unified interface.

The interesting bit Most Go frameworks hand you middleware and routing; Weave treats every capability as an independent plugin with its own namespace and route prefix, then adds dependency resolution and conflict detection so modules don’t step on each other. It’s an operating-system-minded approach to a web backend, which is either over-engineered or exactly what complex AI services need.

Key highlights

• Hot-pluggable plugin system with dependency resolution and conflict detection; all plugins implement the same Plugin interface.
• AI stack covers multi-model LLM invocation (OpenAI, Ollama, ModelScope), AIChat with streaming responses, Agent tool-calling and task planning, and RAG via RedisSearch vector search.
• Security and observability are infrastructure-layer concerns: JWT auth, CSRF protection, token-bucket rate limiting, plus Prometheus/Grafana dashboards and structured zap logging.
• Includes a scaffolding tool to generate plugin boilerplate, and the layered architecture (interface, business, data, infrastructure) keeps the core minimal.

Caveats

• The README touts “high performance” and “excellent performance” but supplies no benchmarks, latency numbers, or load-test results to support the claims.
• The provided source is truncated mid-directory-tree, so details on plugin APIs and advanced configuration remain incomplete.

Verdict Worth evaluating if you want a Go backend that treats LLMs, agents, and RAG as first-class, swappable modules rather than hardcoded dependencies. Skip it if you need a battle-proven framework with extensive production testimonials; at 501 stars it is still proving its case.