Ecosystem maps
The local AI ecosystem moves weekly. These maps are the version of it we update monthly — categorized, captioned, and editorialized so you can see the shape of the territory before drilling into any one entity.
The local AI agent ecosystem (May 2026)
Coding agents, orchestrators, memory frameworks, and inference protocols structured into one map. Where every entry links into a catalog entity, system guide, or both.
The MCP server ecosystem (May 2026)
Anthropic reference servers, vendor-maintained servers (GitHub, Microsoft Playwright), community standouts, the clients that consume them, and the remote-MCP wave.
The local AI inference runtime landscape (May 2026)
Six zones covering every runtime that hosts LLM weights — desktop locals (Ollama/llama.cpp), production servers (vLLM/SGLang/TensorRT-LLM), Apple Silicon (MLX), quantized engines (ExLlamaV2/TabbyAPI), distributed (Exo/Petals/Ray Serve), and agent bridges.
The memory framework ecosystem (May 2026)
Six zones covering every layer of agent memory: vector databases (Chroma / Qdrant / LanceDB / Milvus / Weaviate / Redis), graph memory (Neo4j / Graphiti / Zep), agent frameworks (Letta / Mem0), MCP memory servers, local RAG frontends (AnythingLLM / Open WebUI), and observability tools.
The coding-agent ecosystem (May 2026)
Six zones covering every coding-agent paradigm: autonomous open-source (OpenHands / OpenClaw / Goose), surgical-edit CLI (Aider), IDE-resident (Cline / Continue), closed-source flagships (Claude Code / Cursor), MCP-native infrastructure, and shutdown / migration tracker.