---
title: "Basic Memory vs Hjarni vs MDflow (2026)"
description: "A focused three-way comparison of the markdown-native MCP knowledge bases — Basic Memory, Hjarni, and MDflow — with a decision framework for picking the right one for you."
author: "MDflow"
date: 2026-07-12
reading_time: "13 min"
canonical_url: https://mdflow.cz/blog/basic-memory-vs-hjarni-vs-mdflow
md_url: https://mdflow.cz/blog/basic-memory-vs-hjarni-vs-mdflow.md
---

# Basic Memory vs Hjarni vs MDflow (2026)

*Published July 12, 2026 · 13 min read*


If you have narrowed your search for an AI knowledge base down to **Basic Memory, Hjarni, and MDflow**, you have good taste — these three are the markdown-native options in a category otherwise full of proprietary memory stores. They share a design philosophy most tools don't: your knowledge lives as **plain Markdown files you own**, and a built-in **[Model Context Protocol](/docs/mcp) (MCP) server** lets ChatGPT, Claude, and other AI clients read *and* write them directly. That shared foundation is exactly why they're hard to tell apart from the outside.

This is a focused, three-way comparison to help you pick. We build one of them ([MDflow](/)), so read this as a vendor's honest map of a category we compete in — the trade-offs are real and we've tried to name the ones that favor the other two. If you want the wider survey that also covers non-markdown tools like mem0 and Obsidian, we wrote a separate [best MCP knowledge base](/blog/best-mcp-knowledge-base) roundup. This post is the head-to-head for people who've already decided they want **markdown**.

> **TL;DR** — All three store plain Markdown and ship an MCP server, so the decision isn't "which stores my notes better" — it's about three axes. **Self-hosted or hosted?** Basic Memory is a free, open-source engine you run locally (with a paid Cloud tier); Hjarni and MDflow are hosted, no install. **AI memory or a real workspace?** Hjarni is a clean, memory-first app; MDflow is a full editor with governance around the notes. **How much governance?** MDflow adds automatic version history on every write, sharing, collections, and client-side encryption. Want free and local → [Basic Memory](https://basicmemory.com/). Want simple hosted memory → [Hjarni](https://hjarni.com/). Want a hosted workspace with guardrails → [MDflow](/).

## What these three have in common

Before the differences, it's worth being precise about the shared core, because it's the reason these three end up on the same shortlist:

- **Plain Markdown you own.** No proprietary block model, no database you can't leave. Your knowledge is `.md` files — portable, greppable, and yours. All three let you export everything as Markdown.
- **A built-in MCP server.** Each ships its own MCP server, so you don't assemble one from community plugins. Point an AI client at it and the assistant queries your notes as a tool.
- **Read *and* write.** This is the dividing line between a knowledge base and a read-only archive. All three let the AI create and update notes, not just search them — so your assistant can capture what it learned and keep the base current.
- **Folder organization with instructions.** Each uses folders, and each lets you attach human-written context — folder descriptions or custom instructions — that the AI reads as guidance.

If any tool in the category lacked those four, it wouldn't be in this comparison. So the real question is what happens *around* that core.

## The three-axis decision

The honest way to choose between markdown MCP knowledge bases is to answer three questions about your situation, not to hunt for a "winner." Here's the shape of it.

| | **Basic Memory** | **Hjarni** | **MDflow** |
| --- | --- | --- | --- |
| **Hosting** | Local-first (free), or Cloud (~$15/mo) | Hosted only | Hosted only |
| **Install to start** | Yes (local process) | None | None |
| **Phone / any device** | Cloud tier only | ✅ built in | ✅ built in |
| **Open source** | ✅ AGPL-3.0 | ❌ | ❌ |
| **Center of gravity** | Local engine + knowledge graph | AI long-term memory | Document workspace + editor |
| **Retrieval** | Vector + hybrid + typed graph | Folder-scoped search | `mdflow_get_context` ranks folder descriptions |
| **Version history** | Your files + Git; cloud audit logs | Hosted memory (no full doc history) | ✅ automatic, every write path, one-click restore |
| **Governance extras** | Audit logs (Cloud) | Folder instructions, roles, public links | Sharing, collections, comments, AES-256 encryption |
| **Free tier** | ✅ (local, unlimited) | ✅ up to 25 notes | ✅ generous free plan |
| **Best for** | Developers who want a free, self-hosted engine | The simplest hosted AI memory | A hosted workspace with governance |

### Axis 1 — Self-hosted or hosted?

**This is the biggest fork, and it maps almost perfectly onto Basic Memory vs the other two.**

[Basic Memory](https://basicmemory.com/) is **local-first and open source** (AGPL-3.0, from Basic Machines). The engine runs on your own machine, your notes are Markdown files on your disk "forever," and it's free. That's a genuinely different value proposition: nothing leaves your hardware, you can read the code, and there's no subscription for the local path. It also ships a paid **Cloud** tier (around $15/month) that adds a hosted database, cross-device sync, and mobile access for people who want hosting without self-managing it.

Hjarni and MDflow are **hosted only.** There's nothing to install — you connect an AI client with one URL and reach the same notes from your laptop or your phone. The trade-off is the mirror image of Basic Memory's: you don't own the server, but you also don't run it.

The practical tell is **device reach.** If you'll only ever work from one machine and you value self-hosting, Basic Memory's free local tier is excellent. The moment you want your notes on your phone or a shared computer with zero setup, you're either paying for Basic Memory Cloud or reaching for a hosted tool that does it by default.

### Axis 2 — AI memory, or a real workspace?

**Once you've decided you want hosted, the split between Hjarni and MDflow is about what the notes are *for*.**

[Hjarni](https://hjarni.com/) is a focused, well-built **AI-memory app.** Its pitch is clean: give ChatGPT and Claude long-term memory as Markdown notes in folders you can open and edit, with a built-in MCP server and one connection URL. You can set **custom AI instructions** on a folder, a team, or your whole account, and export everything as a ZIP. Its free tier is generous (up to 25 notes with full MCP and API access), Pro is $10/month for unlimited notes plus file attachments and public folder links, and Teams is $13/seat/month. If what you want is *hosted memory your AI maintains*, Hjarni is an easy recommendation and does that job with very little friction.

[MDflow](/) is a **document workspace that happens to have an MCP server bolted into its core.** The difference shows up when *you* — the human — sit down to write. MDflow is a genuine editor (split live preview, a real folder tree, workspaces, full-text search) with the AI access layer wrapped around it, not the other way around. So it fits when the notes are things you read and write yourself as much as the AI does: specs, docs, research, a personal wiki — where you want an editor *and* the agent access in one place.

The one-line version: **Hjarni optimizes for AI-maintained memory; MDflow optimizes for a knowledge base humans and AI both author.** Neither is wrong — they're aimed slightly differently.

### Axis 3 — How much governance do you need?

**The moment an AI can *write* to your knowledge base, governance stops being a luxury.** An agent edit is a change you didn't type, so the questions become: can you undo it, control who sees it, and keep the sensitive parts private?

This is where the three diverge most:

- **Basic Memory** leans on the fact that your notes are files. Version history is whatever you get from **Git** or the filesystem, plus **audit logs** on the Cloud tier. That's powerful if you're a developer who already lives in Git — and clunky if you're not.
- **Hjarni** offers folder instructions, roles, shared folders, and public folder links — solid team and sharing controls — but it's memory-first, without full per-document version history.
- **MDflow** ships governance as a first-class layer: **automatic version history on every write path** (editor, API, *and* AI agent) with line diffs and one-click restore, public and private [sharing](/faq), [collections](/faq), anchored comments, and optional **client-side AES-256 encryption**. When an agent writes, every change is reversible without you setting anything up.

If your knowledge base is low-stakes personal memory, minimal governance is fine. If multiple people or multiple agents touch it — or it holds anything sensitive — the guardrails matter, and this is the axis where MDflow is built to pull ahead.

## Which applications benefit most

The three tools sort naturally onto different jobs:

1. **A personal, local-first second brain** — one machine, privacy-maximal, developer-comfortable → **Basic Memory** (free, self-hosted).
2. **Long-term memory for ChatGPT and Claude** — you mostly want your assistant to remember things across sessions, hosted, minimal fuss → **Hjarni**.
3. **A shared or authored knowledge base** — specs, docs, research, or a team wiki that both humans and agents write, from any device, with history and access control → **MDflow**.
4. **Agentic coding memory** — an agent that writes and updates a spec/progress file as it works benefits from reversible writes; MDflow's automatic version history and MDflow's or Basic Memory's write-back both fit, depending on hosted vs local.
5. **A knowledge graph you can traverse** — if typed relations between notes are central to your workflow, Basic Memory's graph model is the specialist here.

## How MDflow fits

We built [MDflow](/) for **axis 2 and axis 3**: the person who wants hosted, plain-Markdown knowledge that ChatGPT and Claude can read and write — with a real editor and real governance around it. Here's what lines up today, stated accurately.

### What lines up today

**A hosted remote MCP server, no install.** MDflow's [MCP server](/docs/mcp) lives at `https://mdflow.cz/api/mcp`. Connect Claude, ChatGPT, Cursor, or Codex with a Personal Access Token — or OAuth where the client supports it — and you read and write your notes from any device, no local process to babysit.

**Curated retrieval, not a raw dump.** [`mdflow_get_context`](/docs/mcp) takes a topic and scores your **[folder descriptions](/blog/folder-descriptions-agent-context)** highest, then names and titles, returning the most relevant Markdown bodies as readable context plus structured JSON. Folders nest to any depth and their descriptions cascade, so an agent reads a labeled chain of context before opening a single document — no vector database to maintain.

**Producers, not just readers.** Through the [MCP server](/docs/mcp) and the [HTTP API](/docs/api), agents create, update, move, organize, and share documents — the [write-back](/blog/ai-agents-write-to-knowledge-base) half of the loop, so your AI keeps the base current.

**Governance built in.** Automatic version history on every write path with line diffs and one-click restore, public and private sharing, collections, anchored comments, and optional client-side AES-256 encryption. This is the layer local files and memory-first apps handle differently — or leave to you.

**Plain Markdown with raw `.md` twins.** Every document is portable Markdown, and any shared link serves a raw `.md` twin with YAML frontmatter over open CORS — an assistant can fetch and cite a document in one request. Export is just files; there's no lock-in.

**Discovery for agents.** MDflow ships an [`llms.txt`](/llms.txt) index, an agent manual, an A2A agent card, and an OpenAPI spec, so assistants can find the surface, not just use it. See [Markdown for AI](/markdown-ai) for the full picture.

### Where we're headed

This is **direction, not a dated commitment.** The shape of our thinking: richer collection-level MCP so an agent can pull a whole cross-linked knowledge set at once; first-class document types and tags aligned with emerging standards like Google's [Open Knowledge Format](/blog/open-knowledge-format-adoption-guide); and agent-assisted enrichment that proposes folder descriptions and cross-links for knowledge you already have. The [Web Clipper](/clipper) already turns web pages into clean Markdown; capturing straight into an agent-ready knowledge base is the natural next step.

## The bottom line

There is no single winner among Basic Memory, Hjarni, and MDflow — there's the right one *for your constraint*, and all three are honestly good at what they aim for. Want **free and local**, and you're comfortable self-hosting? [Basic Memory](https://basicmemory.com/) is the strongest open-source engine in the category. Want the **simplest hosted AI memory** with a generous free tier? [Hjarni](https://hjarni.com/) does that cleanly. Want a **hosted workspace** that's also a real editor with version history, sharing, and encryption around notes both you and your AI write? That's the gap [MDflow](/) was built to fill.

Whichever you choose, you've already made the important decision — plain Markdown you own, read and written by your AI over an open protocol, beats re-explaining yourself to ChatGPT every morning.

[Start free](/login) · [Connect an AI agent](/docs/mcp) · [Read the API docs](/docs/api)

## Frequently asked questions

### What do Basic Memory, Hjarni, and MDflow have in common?

All three are markdown-native knowledge bases with a built-in MCP server, so ChatGPT, Claude, and other AI clients can read and write your notes as plain Markdown files you own. That shared core is why they compete for the same user. The differences are about deployment (self-hosted vs hosted), focus (AI memory vs full document workspace), and how much governance ships around the notes.

### Which is the best markdown MCP knowledge base for me?

Pick by your constraint. Choose Basic Memory if you want a free, open-source engine that runs locally on your own disk and you are comfortable self-hosting. Choose Hjarni if you want the simplest hosted AI-memory app and a generous free tier. Choose MDflow if you want a hosted workspace that is also a real editor with governance — automatic version history on every write, sharing, collections, and client-side encryption — reachable from any device including your phone.

### Is Basic Memory free?

Yes, the Basic Memory engine is open source (AGPL-3.0) and free to run locally on your own machine — your notes are plain Markdown on your disk. There is also a paid Cloud tier (around $15/month) that adds a hosted database, cross-device sync, and mobile access without self-managing a server. The trade-off with the free tier is that reaching your notes from a phone or a shared device means running your own sync or paying for Cloud.

### Do any of these run without a local server?

Hjarni and MDflow are hosted, so there is nothing to install — you connect ChatGPT or Claude with one URL (a token or OAuth) and reach the same notes from any device. Basic Memory is local-first by default, so its free tier needs a process running on your machine; its paid Cloud tier is the hosted option. If phone access with zero setup is the priority, a hosted tool is the shorter path.

### Why does version history matter for an AI knowledge base?

Once an AI agent can write to your notes — not just read them — every edit is a change you did not personally type, so you need a safety net. Automatic version history means each write is snapshotted and reversible, so a bad agent edit is one click to restore. MDflow captures version history on every write path (editor, API, and agent); Basic Memory relies on your files plus Git or its cloud audit logs; Hjarni is hosted memory without full document version history.

## Further reading

- Basic Memory — [Open-source AI-native knowledge base](https://basicmemory.com/) · [GitHub](https://github.com/basicmachines-co/basic-memory)
- Hjarni — [Long-term memory for ChatGPT & Claude](https://hjarni.com/)
- Anthropic — [Introducing the Model Context Protocol](https://www.anthropic.com/news/model-context-protocol)
- MDflow — [The best MCP knowledge base (wider roundup)](/blog/best-mcp-knowledge-base) · [Folder descriptions as agent context](/blog/folder-descriptions-agent-context) · [MCP documentation](/docs/mcp) · [API documentation](/docs/api) · [Markdown for AI](/markdown-ai)