Getting started with Hjarni
Hjarni is a Markdown knowledge base your AI reads and writes. Connect ChatGPT or Claude first, then give it a simple structure to work in: your assistant can build most of it with you.
Quick answer
To get started with Hjarni: connect the AI you already use (ChatGPT, Claude, or any MCP client) to Hjarni's built-in MCP server, then ask it to set up your knowledge base. It asks a few questions, saves your first note as Markdown, and can build a simple folder structure with you. From then on it can search, read, and write those notes in any new chat.
Start with the right mental model
Hjarni stores the knowledge
Your notes live in Hjarni as Markdown, with containers, tags, links, instructions, and files.
ChatGPT and Claude are interfaces
They read and write through MCP. They do not become your storage layer or your permanent memory by themselves.
That distinction matters. If something should be available later, save it as a note. If it should shape future AI behavior, save it as an instruction.
Your first 10 minutes
- Connect ChatGPT or Claude. If Hjarni shows a first-note prompt, copy it and paste it into your assistant after connecting. Otherwise ask your assistant to set up your Hjarni. It asks a few questions and saves your first note.
- Create 3 to 5 top-level folders for how you actually work, not how you wish you worked. Your assistant can create them for you.
- Add one note to each important area instead of building a huge folder tree first.
- Pick a small set of reusable tags like
idea,meeting,research, ordraft. - Add one short AI instruction for your most important folder.
Connect the assistant you already use
ChatGPT
Install Hjarni from the ChatGPT Apps Directory in one click, or add it by hand in developer mode. Both end in the same OAuth sign-in.
Claude
Install Hjarni from the Claude Connectors Directory for Claude.ai, Claude iOS and Claude Desktop. Claude Code connects with one claude mcp add command instead.
If you're building your own integration instead of connecting an assistant, skip the setup guides and go straight to the REST API or MCP reference.
Good defaults that age well
Use containers for where a note belongs
Projects, areas, teams, and long-lived domains should usually be containers.
Use tags sparingly and consistently
Statuses and recurring themes make good tags. Avoid creating a new tag for every tiny variation.
Link notes when they are meaningfully related
Links help your AI trace decisions, meetings, research, and source material across containers.
Keep instructions short and operational
Tell the assistant what to include, how to format, and which tags or links matter.
Common mistakes to avoid
- Creating a deeply nested folder tree before you have actual notes to place in it.
- Using tags and containers interchangeably.
- Expecting the assistant to remember things you never saved to Hjarni.
- Writing vague instructions like “organize things well” instead of concrete rules.
Related docs
Need help getting set up?
Email evert@hjarni.com
Give your AI a memory. Free.
Connect Claude or ChatGPT to notes they can actually read and write.
Give your AI a memory. Free.