Your notes and your AI,
finally connected.
You take notes. You journal. You save ideas. Then you talk to your AI and it knows none of it. Grove connects them — your AI reads your notes, writes back, and the knowledge grows together.
The problem
Your notes live in one world. Your AI lives in another.
You have an Obsidian vault, a folder of markdown, years of journal entries. Ideas, people, projects — all written down, some of it connected. That's one world.
Then you open Claude on your phone. Or ChatGPT on the web. Or Cursor in your editor. It has no idea what you've been thinking about. Every conversation starts from zero. That's the other world.
They should be the same world.
How it works
Point Grove at your Obsidian vault, markdown folder, or git repo. It indexes everything — your notes, your frontmatter, your wikilinks — and gives you an MCP endpoint. Add that URL to Claude, ChatGPT, Cursor, or any MCP client.
Ask a question and your AI searches your vault — keyword and semantic search fused together in under 30ms. It finds the right notes, follows the links between them, and answers with your own context.
"What do I know about design systems?"
found 3 concept notes, 2 journal entries, linked to 14 other ideas in your vault — 23ms
When you discover something new in a conversation, your AI can save it directly to your vault. New concepts, updated notes, extracted entities. Every write is validated against your vault's structure and committed to git.
+ Concepts/Context Engineering.md
~ Concepts/Design Systems.md ← new link added
git commit: grove (claude-ai): create Context Engineering
Next conversation — any device, any client — your AI already knows. The note exists. The links are wired. You never explain the same thing twice. The more you use it, the more it knows.
What grows
Not a folder of files. A graph of connected ideas.
Your vault
├── concepts
│ ├── Design Systems ──── linked to 42 other notes
│ ├── Risk Tolerance ──── linked to Financial Plan, journal entries
│ └── Resilience ───────── bridges architecture and personal notes
├── people
│ └── each linked to concepts, projects, conversations
├── journal entries
│ └── entities extracted, wikilinks wired automatically
└── 4,500+ connections
└── the graph sees what folders can'tYour notes aren't isolated files. They're nodes in a graph. A concept about resilience might link your engineering notes to your personal journal — because the same idea shows up in both places. Your AI can follow those connections.
Every note has structured frontmatter. Every link is a real relationship. Grove understands your vault's conventions — types, tags, folder structure — and enforces them on every write. Agents can't corrupt what you've built.
Coming soon
Share a grove — a window into your knowledge.
A grove is a topic-scoped slice of your vault with permission controls. Share your research notes with a colleague. Your recipes with a friend. Your architecture decisions with your team.
They connect their AI to your grove. An LLM judge filters every response server-side — they see exactly what you choose to share, nothing more. Your private notes stay private. The boundaries are enforced by the server, not by trust.
Coming soon
The graph grows while you sleep.
A background loop watches for new and changed notes. It extracts concepts, identifies people, wires links between related ideas, and surfaces surprising connections — all automatically.
Save a bookmark. Write a journal entry. Drop a file in your vault. By the next morning, it's been integrated into the graph, linked to what's already there, ready for your AI to use.
Every autonomous action is a git commit with a blast radius limit. Rollback to any point. The vault is sacred — discovery is a careful gardener, not an unsupervised lawnmower.
Your vault, your choice
Hosted
Connect your GitHub repo. Get an MCP endpoint. We handle search, embeddings, and infrastructure.
Get early access →Self-hosted
Your server, your data. MIT licensed. Self-hosted embeddings — nothing leaves your infrastructure.
View on GitHub →Same API, same MCP protocol across all modes. Works with Claude, ChatGPT, Cursor, and any MCP client.
"I think there is room here for an incredible new product instead of a hacky collection of scripts."
Karpathy described a workflow where LLMs compile raw documents into structured wikis. That works for research collections. But most people don't start with a pile of papers — they start with notes they've already been taking.
There are 24 Obsidian MCP servers on the registry. Every one is local-only, read-only, and treats your vault as a bag of text files. They work from your laptop. Open Claude on your phone — nothing.
Grove is remote, bidirectional, and vault-aware. It doesn't just search your notes — it understands how they connect, writes back with validation, and works from every device you own.
Your AI should know what you know.
Connect your notes. Every conversation gets better.
Open source · MIT licensed · Privacy-first · Git-native