Skip to content

The knowledge base for ChatGPT & Claude

Give ChatGPT & Claude long-term memory.

A hosted Markdown knowledge base your AI can read and write over MCP. Save notes once. Stop re-explaining yourself.

Give your AI a memory See how it works

Free to start. No credit card required.

Available in the ChatGPT Apps Directory We don't train on your data Export to Markdown anytime
Claude
What did we decide about the deadline on Tuesday?
hjarni: search
✓
From your meeting notes: you moved it to March 15 to give the team an extra week. Sam is updating the client.
Change of plan: the client moved it to March 22.
hjarni: notes-update
✓
Updated your meeting note: the deadline is now March 22. The next conversation already knows.
Why did we switch from REST to gRPC for the internal APIs?
hjarni: notes-get
✓
Your notes from the tech review: latency on the order service was 3x worse with JSON serialization. The team benchmarked protobuf and it cut p99 from 120ms to 40ms.
Which service migrated first?
Order service in Q1. Payments and inventory followed in Q2 once the proto schema stabilized.
What did the Series A investors say about our margins?
hjarni: search
✓
From the Sequoia meeting: they wanted gross margins above 70% before raising. You're at 64%.
Draft a one-line answer for the deck.
“Path to 72% gross by Q3 via infra consolidation and the May pricing tier rollout.”
What protocol did the Chen lab use for their cell viability assay?
hjarni: search
✓
From your literature notes: Chen et al. used a flow cytometry protocol with 48-hour incubation. 92% viability at the target concentration.
Did anyone replicate it?
Patel lab hit 89% with the same protocol last March. Your notes flagged their lower seeding density as the likely reason for the gap.
What was the name of the fisherman I interviewed in Galveston?
hjarni: search
✓
Your interview notes from March 12: Ray Guidry, 34 years shrimping the Gulf. He said the catch is half what it was in the '90s.
Pull his best quote for the lede.
“My granddaddy could fill the boat by lunch. Now I’m lucky to fill it by sundown.”
I'm new here. How do we deploy to production?
hjarni: search (Runbooks)
✓
From the team's Deploy runbook: merge to main, CI runs, then run bin/deploy production. It tags the release and runs migrations before the cutover. Deploys are paused on Fridays.
What if a migration fails halfway?
The runbook says roll back with bin/deploy rollback, then post in the incident channel. The Decisions folder notes why migrations are kept backward compatible for one release.
See all use cases

5,000+ notes and counting, kept current for ChatGPT, Claude, and every MCP client.

The problem

Sound familiar?

ChatGPT and Claude start every session from zero

You explained it last week. This week's chat knows none of it. So you explain it again.

Your knowledge is scattered

Notes in one app, docs in another, the detail you need buried in an old chat. Your AI sees none of it.

Your tools do too much

Notion has 47 features you never touch. You just want to write something down and have your AI remember it.

How it works

Three steps. Two minutes.

No new habits. Your AI does the reading and writing.

1

Write or clip a note

Plain Markdown. Project decisions, customer interviews, trip notes. Anything you'd otherwise re-explain. Clip from the web with the Chrome extension.

2

Connect ChatGPT or Claude

One-click from the ChatGPT Apps Directory or OAuth into Claude. Add Perplexity with a custom connector, or Cursor and other tools in two lines of config.

3

Your AI reads and updates

It searches your notes when you ask a question, and writes back what it learned, so the next conversation already knows it. You stop re-pasting context.

A note titled 'Website launch: plan and decisions' open in Hjarni, with planning and launch tags, a one-line summary, and a Markdown body listing decisions and action items. The sidebar shows the Work > Projects folder it lives in.
One note in Hjarni. Plain Markdown, organized in folders, ready for your AI to read.

Maintained, not just saved

Built-in AI memory only remembers. Hjarni keeps it true.

Facts change. You switch jobs, reverse a decision, raise a price. Memory that only appends goes stale. In Hjarni, Claude and ChatGPT find the note that is now wrong and fix it.

Updating beats creating.

notes-update is the most called tool on Hjarni's built-in MCP server. More calls than notes-create. In real use, your AI spends more time correcting and merging notes than writing new ones. That is the difference between a pile and a memory.

1

One fact, one place.

Say what changed. Your AI rewrites the note that says otherwise. No second note contradicting the first.

2

Every edit is visible.

Notes show what changed and who changed it, you or your AI. Review it, or change it back.

3

Better in month six than week one.

Long-time users edit old notes as often as they write new ones. Stored memory piles up. Maintained memory pays off.

See how it compares to ChatGPT's built-in memory

Quick update: I left Acme. I'm head of growth at Vektor now.
hjarni: search
✓
hjarni: notes-update
✓

About me › Work

Updated just now

Work and role

Marketing lead at Acme.

Head of growth at Vektor.

Prefers async updates. Mentors two newer teammates.

Created in January. Merged a duplicate in March. Corrected by Claude today.

One note, kept current. Not three notes that disagree.

The idea

Memory you search, not memory you carry.

A knowledge base you build once and keep for good. Every note makes your AI sharper about your decisions, your standards, and the way you work.

Compounds over time One memory for every AI Yours in plain Markdown

What makes Hjarni different

Your AI works the way you work.

Write the rules on a folder once. Claude and ChatGPT read them before they touch a note. No prompting, no copy-paste.

Your brain

Rules stack on the way down

  • Personal brain-level

    "Prefer bullets over paragraphs. Lowercase titles."

  • Projects

    "Open with the current status and the deadline."

  • Launch you're here

    "List the next step and who owns it."

Claude and ChatGPT read all three when they open a note in Launch.

How folder instructions work

From the people who use it

Hjarni ran the whole Labeur launch. Scope decisions, landing page copy, App Store submission, launch day checklist, all in one place and all in sync with Claude. I stopped re-explaining myself to the AI. From first sketch to live in the store.
Joren De Groof Not A Tech Company
What sold me was how light and well defined the MCP is. My AI used to choke when I pasted Markdown and config files into a chat. It reads them perfectly from Hjarni.
Jack GTM Strategy Consultant

Features

Simple on purpose.

Less software. More thinking.

Built for ChatGPT and Claude

Full REST API and built-in MCP server. Your AI reads and writes your notes directly. Set up ChatGPT or Claude in minutes.

Also works with Cursor, Claude Code, Windsurf, and Cline.

Markdown, end to end

Notes are plain Markdown, the format your AI already speaks. No conversion. No lock-in.

See and undo AI edits

Every edit is saved and attributed to you or your AI. Revert any version in one click, and recover deleted notes from a 30-day Trash. How history works

Connect your thinking

Wiki-links between notes. Build a web of knowledge, not a pile.

Also in the box

Fast inbox capture Web clipper Public folder sharing Export to Markdown Zapier automation Notes, folders, tags Full-text search

Templates

Don't start from a blank page.

Each template is a ready-made folder tree with AI instructions already written. Pick one and your AI knows how to use it from the first note.

Developer Project Memory

A ready-made knowledge base for your codebase. Architecture, conventions, decisions, runbooks. Claude and ChatGPT read it before they write a line.

Use this template

Customer Interview Repository

A shape for the interviews that matter. Save the raw notes, tag the themes, and let Claude or ChatGPT do the synthesis from real material. Stop re-reading the same transcripts.

Use this template

Meeting Memory

A folder structure for the meetings that matter. Decisions, follow-ups, attendees. Claude and ChatGPT read it the next time anyone asks "what did we decide last week".

Use this template

Writing Style Guide

A style guide your AI reads before it drafts. Voice, banned words and habits, patterns, real examples. The draft starts closer to your voice, not generic AI default.

Use this template

Reading List

Organize books and articles you want to read, are currently reading, and have finished. Your AI assistant will track progress and collect your notes and highlights.

Use this template

Travel Profile

Your travel taste in writing. Preferences, favorite places, past trips, wishlist. Claude and ChatGPT plan trips that feel like you, not a generic itinerary.

Use this template

Browse all templates

Teams

One person writes it down. Everyone's AI knows it.

Share a folder. Every teammate's AI reads from the same notes.

A shared Hjarni folder with notes written by four different teammates

Write once, the whole team remembers

One person writes down how it works. Now everyone's AI knows it. No more "did you see that doc?"

Set rules once, the whole team follows

Write down how your team does things once. Everyone's AI follows it, new people included. No more starting from zero.

Start your team free See Hjarni for teams

Building one for a whole company? See the company brain.

Pricing

Simple pricing

Free plan forever. No credit card. No hidden fees.

Free

€0 forever

Up to 25 notes with full MCP & API access. Enough to see your AI remember.

Pro Most popular

€9 /month

Unlimited notes, file attachments, and sharing. Personal long-term memory for ChatGPT and Claude.

Teams

€18 /seat/month

Shared memory for a team's AI tools. Free for your first 25 team notes.

Start writing for free Compare plans in detail

2 months free on annual billing. Cancel anytime, you keep every note.

  • EU-hosted (Hetzner)
  • Encrypted in transit & at rest
  • We never train on your notes
  • Export to Markdown anytime
  • GDPR-aligned
Security & privacy

Common questions

Is this just another note-taking app?

No. It's a knowledge base your AI can actually use. Your notes become live context for every AI conversation.

ChatGPT already has memory. Why do I need this?

Built-in memory only remembers. It lives inside one app, collects facts as it goes, and grows stale when something changes. Hjarni is notes you own and edit, readable by ChatGPT, Claude, and any MCP client. When a fact changes, your AI updates the note that says otherwise. See the full comparison: Hjarni vs ChatGPT memory.

Why not Notion or Obsidian?

Great tools. Different goals. Hjarni is built from the ground up for AI. Custom instructions per folder. Team-wide AI conventions. An MCP server designed around your knowledge, not bolted onto a workspace. See how Hjarni compares to Notion, Obsidian, ChatGPT Projects, Claude Projects, and Apple Notes.

What's MCP?

Model Context Protocol. An open standard that lets AI assistants connect to external tools and data. Hjarni has a built-in MCP server. Your AI reads and writes your notes directly.

Which AI assistants work with Hjarni?

ChatGPT and Claude connect via MCP today. Any AI that supports MCP will work automatically. Others can use the REST API.

How do I give my AI a memory?

Add Hjarni as an MCP server in ChatGPT or Claude. Takes two minutes. Every conversation picks up where the last one left off.

What happens to my data?

Your notes live on our servers. Export everything as Markdown anytime. We don't train AI on your data. We don't sell it. It's yours.

Can I collaborate with others?

Yes. Create a team, invite members, share folders. Everyone's AI benefits from the shared knowledge.

Can I self-host Hjarni?

Not currently. It's a hosted service. That keeps things simple, for you and for us.

Start here

Write once. You both remember.

Free to start. No credit card required.

Give your AI a memory

Works with Claude and ChatGPT today.