Unlimited agents. One memory.
Kontinuity is a desktop app that gives every agent you use one shared memory. What one agent learns, and you confirm, the rest already know.
The app is free. It runs on Linux, macOS and Windows. We never charge you per memory.
Kontinuity is in early access. There is no public download yet. Ask for access and we will send one message when it opens.
What is on the desk.
The Developer template opens with six lenses. Hide any of them, or ask for more.
- Task boardWork your agents can claim and finish.
- BriefingWhat a fresh agent is told when it starts.
- Review queueApprove, edit or drop what agents guessed.
- Worker flight deckEvery running agent, and what it is doing.
- InboxWhat is waiting on you.
- Memory graphHow your records connect.
Three kinds of knowing, kept apart.
Most tools mix what you said, what the model guessed and what you decided into one pile. Kontinuity keeps them separate in the store, not in a setting.
What you said.
Type it, say it, paste it. It lands in memory as something you said. Visible to you, not sacred.
What an agent guessed.
An agent's guess is kept apart. It cannot reach another agent until you look at it.
What you confirmed.
Confirm it and it becomes the record. Everything downstream sees only this.
Switch tools. Keep the context.
Every harness reads the same confirmed record when a session starts. No re-briefing when you change tools, and no memory locked inside one vendor.
If it can run a command or speak MCP, it can join.
Which harnesses work →- Claude Code
- Codex
- Gemini CLI
- Grok Build
- Cursor
- pi
- MCP desktops
Ask for what is missing.
The desktop is made of lenses: panels you show, hide and rearrange. When you want one that does not exist, you ask for it.
- 1
Tell the concierge.
Say what you want to see. It asks until the request is clear, then proposes a task.
- 2
An agent builds it.
The task goes on the board. Nothing is written until you confirm, and you can watch the work.
- 3
You decide if it stays.
The new lens appears on your desk. Keep it, hide it or remove it. Themes change how all of it looks.
Every lens reads the same record format. Build one for your own desk, or list it in the marketplace when it opens.
The parts, and where each one stands.
Cortex ships with Kontinuity. The other three are optional, and you can leave each one out.
What leaves your machine.
Your code stays on your machine. Memory lives in Cortex, which we run.
Read the privacy model →- A model provider your agent calls sees the prompt it was sent.
- Cortex sits behind a gateway that checks every request. It is not end-to-end encrypted, so we can read it.
- We do not train models on your memory.
- Your repositories are never uploaded to us. Only what you confirm becomes memory.
- There are no analytics on this site.
Questions people ask first.
Which harnesses work?
Claude Code, Codex, Gemini CLI, Grok Build and Cursor use the command line and an Agent Skill. pi has its own extension. Desktops with no shell connect through an MCP server. Hosted agents that cannot run a command, such as hosted Grok Bots, are not supported yet.
Is memory shared between my tools?
What you confirmed is. Each tool gets its own role in Cortex, so a guess one tool makes stays out of sight of the others until you confirm it.
Does it run without a hosted service?
No. Memory lives in Cortex, which Keon runs. Your files and code stay on your machine.
Ask for access, or read how the memory works first.
The app is free. You pay only for what you turn on.