MD Chat

Your .md file IS the chat history.

Chat with AI directly inside Markdown files. No side panels, no separate apps — just write ## User, hit Ctrl+Enter, and the response streams right back into your document.

Why MD Chat?

• File-first — prompts, responses, notes, and code all live in one .md file. Easy to diff, search, share, and version control.
• 6 providers — Claude CLI, Codex CLI, OpenAI, Anthropic API, OpenRouter, Ollama. Use cloud APIs or run local models.
• Streaming — AI replies stream into the document in real time. Cancel anytime with Escape.
• Session continuity — CLI providers auto-save session IDs in frontmatter. Pick up where you left off.
• @[read] context — inline file contents into your prompt with @[read](https://github.com/Ayjc/md-chat/blob/HEAD/src/main.ts). The AI sees the file, your .md stays clean.
• Extract & Fork — pull code blocks into real files, or branch a conversation into a new chat.
• Multi-file chats — run separate conversations in different files at the same time.
• Per-file overrides — provider, model, system prompt, temperature — all configurable per file via frontmatter.

Quick Start

Option A: Use a CLI provider (no API key needed)

1. Install Claude Code or Codex CLI
2. Open any .md file and write:

## User
What is the capital of France?

3. Press Ctrl+Enter — the response streams in as ## Assistant

Option B: Use an API provider

1. Open VS Code Settings and search for mdChat
2. Set your API key (e.g. mdChat.providers.openai.apiKey)
3. Set mdChat.defaultProvider to openai (or anthropic, openrouter, ollama)
4. Write ## User, press Ctrl+Enter

That's it. Keep adding ## User blocks to continue the conversation.

Providers

Fallback chain — if the default provider is unavailable, MD Chat automatically tries the next one in mdChat.fallbackProviders.

Per-file override — set provider: openai in frontmatter to use a different provider for a specific file.

Frontmatter

Control behavior per file with YAML frontmatter:

---
provider: openai
model: gpt-4o
temperature: 0.7
systemPrompt: You are a concise technical writer.
---

## User
Explain this architecture diagram.

![diagram](https://github.com/Ayjc/md-chat/raw/HEAD/images/arch.png)

@[read] — Inline File Context

Reference source files directly in your prompt. The file content is sent to the AI, but your .md file stays clean:

## User
Please review this file:
@[read](https://github.com/Ayjc/md-chat/blob/HEAD/src/utils/config.ts)

Focus on lines 10-25:
@[read](https://github.com/Ayjc/md-chat/blob/HEAD/src/utils/config.ts#L10-L25)

The AI receives the actual file contents wrapped in a code fence with language detection. Your .md file keeps the @[read] syntax as-is.

Syntax:

• @[read](https://github.com/Ayjc/md-chat/blob/HEAD/path) — read the entire file
• @[read](https://github.com/Ayjc/md-chat/blob/HEAD/path#L10-L25) — read lines 10 through 25
• @[read](https://github.com/Ayjc/md-chat/blob/HEAD/path#L42) — read a single line
• Paths are relative to the .md file's directory

Commands

Extract Code Block

When the AI gives you a code block, press Ctrl+Shift+E to save it to a file. The file extension is auto-detected from the language label (e.g. ```typescript → .ts).

Fork Chat

Press Ctrl+Shift+F to branch the conversation at the current position. Creates a new file with the history up to that point, clears the session, and adds an empty ## User section — ready for a different direction.

Settings

General

CLI Providers

API Providers

Requirements

• VS Code 1.85+
• At least one of:
  • Claude Code (claude on PATH)
  • Codex CLI (codex on PATH)
  • An API key for OpenAI, Anthropic, or OpenRouter
  • Ollama running locally

License

MIT