AgentLLM 1.0.0
Generic LLM provider framework for macOS apps.
Types
- LLMProvider — Protocol for any LLM (sendStreaming, fetchModels)
- LLMProviderConfig — Full config (endpoint, auth, model, capabilities)
- LLMEndpoint — Connection details (chat URL, models URL, auth, headers, port)
- LLMProviderKind — Hosting type (cloudAPI, localServer, remoteServer, embedded, custom)
- LLMAPIProtocol — API format (anthropic, openAI, ollama, foundationModel, custom)
- LLMCapability — Feature flags (streaming, tools, vision, caching, thinking, webSearch)
- LLMModelInfo — Model metadata (id, name, context window)
- LLMResponse — Standardized response (content, stop reason, tokens)
- LLMRegistry — O(1) provider registry
Design
- Zero hardcoded providers — app owns all config
- Protocol-based — implement LLMProvider to add any LLM
- Well-known ports: Ollama=11434, LM Studio=1234, vLLM=8000
- Multi-protocol support (e.g. LM Studio: OpenAI, Anthropic, Native)