Typora Plugin Chat

English | 中文

An AI chat assistant for Typora built on top of typora-community-plugin.

Embedded in Typora, the AI chat assistant can access your vault: read, edit, create and delete files; fetch web content; run sub-agents for delegated tasks; and use reusable skills.

WARNING

LLM automated operations may accidentally modify or delete your files. Please back up your versions in advance (e.g., using git).

Preview

Features

Chat & Conversation

• AI Chat Panel — A full conversational UI in Typora's right sidebar with an auto-resizing input box, model/mode pickers, and slash command autocomplete
• Multi-session Management — Create, switch between, and delete independent chat sessions; transcripts persist across restarts; sessions are scoped per vault
• "Chat About This Note" — Send the current note's content to the AI for summarization or discussion (also available via right-click in file explorer)
• "Send Selection to Chat" — Send selected text into the chat with a scope constraint ("work only within this selection")

Agent Capabilities

• 3 Access Modes: Workspace Write (full access), Read Only, Chat Only (no file tools)
• 17+ Tools across 4 categories: Editor, Vault, System, and Network
• Sub-Agent System — Delegate independent tasks to nested agent loops with inherited permissions and turn-scoped undo/redo
• Skills System — Auto-discover and invoke reusable SKILL.md instruction sets from your vault or global directories
• AGENTS.md Support — The vault-root AGENTS.md is injected into the system prompt so project conventions apply to every agent turn (main loop + sub-agents)

Smart Features

• Slash Commands: /new, /sessions, /skill:<name>, /undo, /redo with keyboard-navigable autocomplete
• Undo/Redo — Turn-scoped atomic undo within the current session: all file changes from one agent turn are rolled back together
• Context Window Usage Ring — Real-time visual indicator of token usage vs. context window size (estimated or API-reported)

Tool Enhancements

• Pagination Support — list_files, web_fetch and other tools support limit/offset pagination to avoid returning too much content at once
• Forbidden Path Policy — Configure glob patterns to block the agent from accessing sensitive directories and files (.git, node_modules, *.apikey are always forbidden)
• Directories in list_files — Non-recursive list_files returns both files and directories

AI Providers

• Anthropic (Claude) — /v1/messages endpoint with tool use, thinking blocks, and token reporting
• OpenAI (GPT) — Responses API (default) or Chat Completions fallback mode
• Alibaba (Qwen) / DeepSeek / Google (Gemini) / MiniMax / Moonshot (Kimi) / SpaceXAI (Grok) / Xiaomi (MiMo) / Z.AI (GLM) / OpenRouter — via standard /v1/chat/completions (OpenAI-compatible)
• Ollama / LM Studio — Local models in the standard /v1/chat/completions format

Settings

Setting	Description
Provider Configuration	Built-in Anthropic, OpenAI, Alibaba (Qwen), DeepSeek, Google (Gemini), MiniMax, Moonshot (Kimi), SpaceXAI (Grok), Xiaomi (MiMo), Z.AI (GLM), OpenRouter, Ollama, and LM Studio; add/remove custom providers. Each provider can be enabled/disabled with independent API Key, Base URL, and model configuration
Model Parameters	Configure parameters such as context window size per model; only models added to the list appear in the sidebar's model picker dropdown
Forbidden Paths	Glob patterns (one per line) to block the agent from accessing specified vault paths (.git, node_modules, *.apikey are always forbidden)
Max Tool Iterations	Max tool-call rounds in one agent loop (default 20, range 1-100)
Skills	SKILL.md files auto-loaded from the specified directories; invoke with /skill:<name> in chat or let the model match automatically

Install

1. Install typora-community-plugin
2. Open "Settings -> Community Plugins", search for "Chat", then install it.