Skip to main content
Enconvo supports local inference through MLX, Ollama, and LM Studio. Choose the runtime that serves the model you need and verify each stage of your task if you want to keep processing on-device.

​
Start small

  1. Configure the local provider and download/load a model.
  2. Send a short text-only request.
  3. Add larger documents or tools after the basic request succeeds.
  4. Check memory use and response time before choosing a larger model.
Model parameter count, quantization, context size, and concurrent tasks all affect memory. Use the model manager’s actual size and compatibility information instead of a fixed RAM table.

​
Keep the whole task local

Review more than the chat model: speech recognition, text-to-speech, OCR, document parsing, embeddings, reranking, search, and MCP tools can use separate providers. Enconvo offers local providers for several of these categories, listed in the catalog. Web search and external services still require network access. A local model can also call tools that send content outside the Mac.

​
LAN connections

When using a server on another computer, configure its LAN address and listen interface. Check macOS Local Network permission if connection errors mention the network being blocked. localhost refers to the Mac running Enconvo, not the remote model server. See Privacy and Security for data-flow checks.