Introducing Kumbukum: Persistent Memory for Every AI Tool You Use
You've used AI tools long enough to know the drill. New chat. Blank slate. Time to re-explain who you are, what you're building, and the same preferences you've already typed out a hundred times in a hundred previous sessions.
That's not an AI problem. That's a memory problem. And it's the exact problem Kumbukum was built for.
What is Kumbukum?
Kumbukum is a persistent memory layer for AI tools. It sits alongside the AI tools you already use, Claude Desktop, Cursor, ChatGPT, Zed, or any MCP-compatible client, and gives them access to a shared, permanent memory store.
Your preferences, project context, past decisions, reference1-million-token context window is useful within a session, but it resets when thenotes, and anything you want your AI to know, you store it once. Every AI tool that connects to Kumbukum can automatically retrieve it. No copy-pasting. No re-explaining. No starting from zero.
The name comes from Kumbukumbu, the Swahili word for memory. We thought that was fitting.
Why We Built It
AI tools have gotten remarkably capable. But they share one fundamental flaw: they're stateless. Each session is isolated. The model doesn't know what you discussed yesterday, which project you're in the middle of, or that you prefer concise answers without bullet points.
The frustration is real. Here are three things we kept hearing:
"I've told Claude my preferences a hundred times. Every session I start over."
"I use three different AI tools. None of them share context. I'm constantly repeating myself."
"I want my AI to remember decisions I've made, not just what we talked about five minutes ago."
The context window isn't the answer either. A 1-million-token context window is useful within a session, but it resets when the session ends. You need something that persists across sessions, across tools, across time.
That's Kumbukum.
Markdown Is Not Memory
Many developers reach for Markdown files when they want their AI to remember something. Drop a CONTEXT.md in the repo, paste your preferences into a README, keep a notes.md updated by hand. It feels like a solution. It isn't.
Markdown is documentation. Memory is something different. A note is something you write down deliberately, a reference you can look up later. Memory is context that gets retrieved automatically, without you having to think about where it lives or whether you remembered to update the file. One is a filing cabinet. The other is recall.
In Kumbukum, both coexist. Notes are longer, structured documents you author intentionally, the kind of thing you'd put in a wiki. Memories are shorter, contextual items the AI stores and surfaces on its own: your preferences, a decision you made last week, the stack your team settled on. You don't manage them manually. They just exist and get pulled in when relevant.
The other problem with Markdown files is that they're local. A file on your laptop isn't visible to Cursor on another machine, or to a team, or to ChatGPT in a browser tab. File systems are not client-server. They don't scale across tools, devices, or people. They were never designed for the multi-tool, multi-session world most AI users live in now.
Tags, Not Folders
Kumbukum doesn't use a file system. There are no folders to create, no directory structures to maintain, no decisions about whether something belongs in /projects/alpha or /context/general.
Instead, Kumbukum uses tags. When your AI stores a memory, it automatically assigns logical tags based on the content. The result is a structured, searchable tree that organizes itself. You can browse by tag or let the AI retrieve what's relevant without thinking about organization at all.
This matters because file systems impose a single hierarchy on information that doesn't naturally fit one. A memory about your coding preferences is relevant to every project. A note about a client belongs in multiple contexts. Tags handle that naturally. Folders don't.
And because it's a cloud-native, client-server architecture, the same tagged memory store is instantly available to every tool you connect, on every device, in every session. No syncing. No copying files around. No need to wonder whether the version on your desktop matches the one on your work machine.
How It Works
Kumbukum is a standalone MCP server. MCP, the Model Context Protocol, is an open standard that lets AI tools connect to external services. Think of it as a plugin system for AI: you add a server URL and your AI gains new capabilities.
Setup takes about 60 seconds:
1. Create a free account at kumbukum.com and get your API key.
2. Paste your MCP server URL into your AI tool's settings.
3. Done. Your AI can now store and retrieve memories across every session.
No code required. If you can follow a three-step tutorial, you can set this up.
What You Can Store
Kumbukum separates two types of stored information:
Memories: short-form context. Preferences, decisions, facts about your work, and things your AI will keep in mind. "The user prefers TypeScript over JavaScript." "They decided to use Postgres, not MySQL." "The client's brand voice is direct and never corporate." You don't need to worry about Memories, as the AI client mostly uses and maintains memories on their own.
Notes: longer, structured documents. Reference material, meeting notes, documentation, anything your AI might need to look up during a session.
Both are searchable using semantic search, so your AI finds the right memory using natural language, not just keyword matching. Fast retrieval even across thousands of stored items.
Which Tools Does It Work With?
Any tool that supports MCP. That includes Claude Desktop, Cursor, ChatGPT (via MCP connector), Zed, OpenClaw, and anything else built on the Model Context Protocol.
One memory store. All your tools. The context travels with you regardless of which AI tool you use.
Pricing: Hosted or Self-Hosted, Your Call
We built Kumbukum for two types of users, and the pricing reflects that.
Free plan: 500 memories and notes, 1,000 API calls per month, one project, semantic search, and all MCP tools. No credit card required. Good for personal use or just trying it out.
Pro plan ($15/month): unlimited memories, notes, projects, and API calls. Flat pricing, no usage anxiety. Priority support included.
One more thing that matters: Kumbukum is open source. You can inspect the code, self-host it, or contribute at the GitHub repository.
No per-call pricing. No usage tiers that punish you for actually using the product.
Is My Data Private?
Yes. Your memories are private to your account. They are never shared with other users and never used to train AI models. You can export or delete your data at any time.
On the Lifetime plan, you can also deploy Kumbukum on your own infrastructure entirely, which means your data never leaves your own servers.
Kumbukum fits naturally into a broader productivity stack. If your team uses Razuna for digital asset management, you can store project context and creative decisions in Kumbukum so every AI session starts with the full picture, not just the files.
If your team runs customer support or internal communication through a shared inbox like Helpmonks, Kumbukum gives your AI the persistent context it needs to handle threads consistently, without re-briefing it every time.
Frequently Asked Questions
What does Kumbukum mean?
Kumbukumbu is the Swahili word for memory. We shortened it to Kumbukum.
Do I need to know how to code to use Kumbukum?
No. Setup is copy-and-paste: you get a URL and an API key, then paste them into your AI tool's settings. Under five minutes.
What is MCP?
MCP stands for Model Context Protocol. It's an open standard that allows AI tools to connect to external services. Kumbukum uses MCP to give your AI access to your personal memory store.
What's the difference between the Pro and Lifetime plan?
The Pro plan ($15/month) is a hosted service: we run the infrastructure, you get unlimited use. The Lifetime plan ($399, one-time) gives you an open-source license, the right to deploy Kumbukum on your own servers, and access to all future updates forever.
We're in early access. The free tier is guaranteed for everyone who signs up before launch.
Try Kumbukum free and stop re-introducing yourself to your AI!