AI Chat and Agent

NoteGen offers two conversational workflows. Chat answers and generates content directly. Agent can plan and call built-in tools, MCP, and Skills to complete multi-step tasks.

Chat

Chat works well for questions, summaries, rewriting, translation, brainstorming, and context-based generation. A request may include:

• current input and conversation history;
• records from a linked tag;
• selected files or folders;
• a quoted editor selection;
• image attachments;
• knowledge-base excerpts;
• memories and Skills matched to the request.

Toolbar controls include model, prompt, response language, linked tags/files, MCP, knowledge base, clipboard monitoring, new chat, and context clearing.

Agent

Agent follows a loop of understanding, planning, tool use, and result checking. Built-in tools can:

• read, create, and modify Markdown;
• create, move, and manage folders;
• query and manage tags, records, and chats;
• save and retrieve long-term memory;
• inspect system and workspace information;
• call currently enabled MCP tools.

During a run, inspect the plan, timeline, tool arguments, and results. A failed run can continue with its context preserved.

Permission and approval

Read-only and mutating operations have different risk levels. Agent displays an approval panel when required. Allow once, approve for the current session, or reject. Inspect targets and arguments before deletion, overwrite, or external-tool calls.

Editor changes appear as a diff before they are written. Accept to apply or reject to keep the original. Multi-file tasks can be reviewed item by item in the changes panel.

Context management

• New Chat creates a separate conversation and keeps the old one in history.
• Clear Context excludes earlier messages from future requests while keeping them visible.
• Clear Chat deletes the current messages.
• Automatic condensation summarizes older messages after configured thresholds.
• Context tray shows linked files, records, memory, knowledge-base context, and tools.

More context is not always better. Reduce linked files, disable unnecessary knowledge-base/MCP sources, or start a new conversation when old information causes confusion or high cost.

MCP, Skills, knowledge base, and memory

These capabilities can work together, but they also expand the data available to the model. Review Privacy and data before handling sensitive content.

Message actions

AI responses can be copied, read aloud, translated, inspected for model details, saved as records, or written to notes. Available buttons depend on toolbar configuration and platform.