MLX-Native Training & DPOLoRA, QLoRA, and DPO preference training on M-series unified memory. Every diff approve/reject becomes training data automatically.
Private InferenceFull-context chat runs offline. Conversations, branching, search, and syntax validation included.
Retrieval-Augmented GenerationIngest documents into local collections. Query them at inference time. Optional web retrieval via DuckDuckGo.
Agent Terminal & SubagentsNanoCore agent with diff proposals, self-healing, and subagent workers. Delegate code review, test writing, docs, and bug fixing to focused workers.
Model Context ProtocolConnect to MCP servers. Discover and execute tools. Generate training datasets from tool schemas.
Data PipelineCSV and JSONL preview. Column mapping, format conversion, PII stripping. Synthetic data generation via MCP.
Model Lifecycle & RoutingPull from Hugging Face. Auto-discover from LM Studio and Ollama. Load, unload, quantize, export. Route different models to different agent roles.