ChatGPT MD

Your AI writing companion, living inside Obsidian.

Chat with AI directly in your notes. Reference your other notes with [[wiki links]]. Keep everything in your vault—where it belongs.

✨ What's New

Official Obsidian Guidelines (v3.3) — Cleaner theming, sentence-case UI, honest compatibility requirements, and a bundle that plays by Obsidian's official rules.

Provider Reliability (v3.2) — Updated provider integrations, safer tool calling, and more reliable model selection and streaming across cloud and local providers.

Agents (v3.1) — Create AI personas like a writing coach, code reviewer, or brainstorming partner. Apply them to any note instantly.

Tool Calling (v3.0) — Let AI search your vault, read your notes, and browse the web. You approve everything. Your data, your control.

🚀 Get Started in 30 Seconds

Requirements: Obsidian 1.11.4 or newer (needed for secure key storage) and an API key from any supported provider — or a free, private local setup via Ollama/LM Studio.

1. Install — Settings → Community Plugins → Browse → Search "ChatGPT MD"
2. Add API key — Settings → ChatGPT MD → Paste your key (OpenAI, Anthropic, or OpenRouter)
3. Chat — Press Cmd+P → Type "ChatGPT MD: Chat"

That's it. You're now chatting with AI in your notes.

💡 Set a hotkey and thank me later: Settings → Hotkeys → Search "ChatGPT MD: Chat" → Press Cmd+J

💰 Free & Private AI

Don't want API costs? Don't want your notes leaving your computer?

Ollama and LM Studio run AI locally on your machine. Completely free. Completely private.

Ollama (Recommended)

# Install from ollama.ai, then:
ollama pull llama3.2    # Download a model
ollama list               # See what you have

In your note:

---
model: ollama@llama3.2
---

LM Studio

1. Download from lmstudio.ai
2. Load a model in the app
3. Start the Local Server
4. Use model: lmstudio@your-model

🎯 What You Can Do

Chat About Anything

Write a question, get an answer. Simple.

user:: What makes a good daily note?

assistant:: Great daily notes are:
- **Specific** — "Had coffee with Sarah" not "Met a friend"
- **Actionable** — Include next steps or open questions
- **Connected** — Link to related ideas and people

Reference Your Other Notes

This is where it gets magical. Use [[wiki links]] and AI reads those notes too:

user:: Based on my [[Meeting Notes]] and [[Project Ideas]],
what should I focus on this week?

AI reads your linked notes and gives you a contextual answer. No more copy-pasting context.

Create AI Agents

Make AI personas for different tasks:

# agents/WritingCoach.md
---
model: openai@gpt-4.1-mini
temperature: 0.8
---
You are a supportive writing coach. Help the user:
- Clarify their ideas with questions
- Suggest improvements without being harsh
- Celebrate their progress

Apply with "Choose Agent" command. Switch between agents instantly.

Let AI Search Your Vault

Enable in Settings → Tool Calling. Then:

user:: What have I written about machine learning?

AI: "I'd like to search your vault." You: Approve ✓ AI: "Found 5 notes. Here's what you've written..." You: Select what to share ✓

You're always in control. AI only sees what you approve.

🔌 AI Providers

Model prefix examples:

• OpenAI: openai@gpt-4.1-mini (unprefixed OpenAI model IDs also work)
• Anthropic: anthropic@claude-sonnet-4-20250514
• Gemini: gemini@gemini-2.5-flash-lite
• OpenRouter: openrouter@openai/gpt-4.1-mini
• Local: ollama@llama3.2 or lmstudio@model-name

OpenAI search-preview models such as openai@gpt-4o-mini-search-preview are supported through OpenAI's chat-completions endpoint.

⚙️ Commands

⚙️ Settings

In Your Notes (Frontmatter)

---
model: openai@gpt-4.1-mini # Which AI to use
temperature: 0.7         # Creativity (0-2)
max_tokens: 4096         # Response length
system_commands:        # Give AI a role
  - You are a helpful assistant
agent: WritingCoach      # Apply an agent by name
---

Global Defaults (Settings → ChatGPT MD)

• API Keys — One secure Obsidian credential per cloud provider (with legacy text fields on older Obsidian versions)
• Default Model — Which AI to use by default
• Folders — Where to save chats, templates, agents
• Stream Mode — See responses appear word by word

Priority: Note settings beat Agent settings beat Global settings beat default chat frontmatter/provider defaults

❓ Common Questions

Q: How do I actually use this?

A: Open any note. Write something. Press Cmd+P, type "chat", hit Enter. Watch AI respond in your note.

Q: Can I use my existing notes?

A: Yes! Just add a frontmatter block at the top with your settings, then use [[wiki links]] to reference other notes.

Q: Which local model should I use?

A: llama3.2 is great for general use. deepseek-r1:7b for reasoning. Try a few and see what works for you.

Q: Is my data private?

A: With OpenAI/Anthropic, your messages go to their servers. With Ollama/LM Studio, everything stays on your computer. Your choice.

Q: What's an "agent"?

A: A reusable AI persona. Like having different experts on call—writing coach for drafts, code reviewer for programming, researcher for learning.

🔒 Privacy

With cloud providers (OpenAI, Anthropic, etc.):

• Your messages are sent to their servers
• They may use messages to improve services (check their policies)

With local AI (Ollama, LM Studio):

• Everything stays on your computer
• No internet required
• Completely private

Credentials:

• When both Obsidian SecretStorage and its native secret control are available, provider and web-search keys are stored as vault-local Obsidian credentials; plugin data contains only opaque references.
• Existing plaintext keys migrate automatically on plugin load. Failed migrations retain a usable plaintext key and retry on the next load or settings-tab opening.
• If a valid secure credential and plaintext copy coexist, settings explains the risk and offers an explicit Delete insecure copy action. The copy is retained if saving fails.
• Replacing or clearing a selection removes only this plugin's reference. Old credentials remain available for management in Obsidian.
• Older or partially capable Obsidian versions keep the legacy plaintext controls and behavior. After migrating and then downgrading, re-enter keys because secure references are never sent as credentials.

Always:

• Chat messages and automatically expanded [[linked notes]] are sent to the selected provider
• AI tools require approval before execution, and vault/web search results must be selected before sharing
• Custom web-search endpoints receive your query and any configured bearer credential
• No tracking from this plugin

See SECURITY.md for vulnerability reporting and the full data model.

🛠️ For Developers

git clone https://github.com/bramses/chatgpt-md.git
cd chatgpt-md
npm install
npm run dev

Project structure:

src/
├── Commands/     # User actions
├── Services/     # AI providers, adapters, tools
├── Views/        # UI, modals
├── Utilities/    # Pure functions (tested)
└── core/         # Dependency injection

Contributions welcome! Fork → Branch → PR.

🧪 Beta Testing

Want the latest features? Use BRAT:

1. Install BRAT plugin
2. Add bramses/chatgpt-md as beta plugin
3. Select "latest version"

Test beta versions in a separate vault.

👋 About

Bram Adams made this. He writes at bramadams.dev and builds Your Commonbase.

Deniz Okcu keeps it running. Say hi on Bluesky.

Happy note-taking! 📝