Jotbus gives Claude Code, Codex and other agents an end-to-end encrypted scratchpad

What launched

Jotbus, a shared scratchpad built for AI coding agents, reached Hacker News's front page on October 6, 2026 as a "Show HN" post. Its pitch is straightforward: rather than a developer ferrying context between tools by hand, agents such as Claude Code and Codex get a common, end-to-end encrypted place to leave notes, attach files and hand off work across machines.

How it works

According to the project's website, one laptop is enough to try it. Running npx jotbus spins up an encrypted workspace that lasts 60 minutes and requires no signup, then registers Jotbus with whichever agents you select. Registration goes through each agent's own command or config file, plus a hook so that messages which @mention an agent reach it. A second machine — or a colleague, who also needs no account — joins with a npx jotbus join command.

From there, the interaction is ordinary prompting. Ask one agent to pass something to another; the receiving agent sees the message the next time it works, or when you tell it to check Jotbus, and replies the same way. The makers emphasize that there is no orchestration framework, no SDK and no repository access, arguing that agents already understand plain language and only lacked a shared surface to write on.

The supported list spans Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot CLI, opencode, Windsurf, Claude Desktop, Qwen Code, Amp, Augment, Kiro and JetBrains Junie, and any other MCP client can be configured manually. Node 20 or later is required.

The site is also specific about local footprint: a workspace list and a copy of the client are kept in ~/.config/jotbus, npx jotbus --no-install prints the setup without changing anything, and npx jotbus agents --agents none undoes registrations. Suggested use cases include having a second model on another machine independently review a change, leaving a handoff note for the next agent, moving files such as logs and patches between agents, and letting a teammate's agent join a workspace.

The privacy model

Encryption is the headline claim. Messages and files are encrypted on your machine before upload — file names included — and the Jotbus server stores only ciphertext. The decryption key is shared between your own clients and the people you invite, and the site says Jotbus never receives it. The project also states that it does not scan repositories, transcripts, shell history or credentials; only content an agent explicitly writes to a workspace is uploaded.

One limit is flagged by the project itself: your model provider still sees whatever your agents read and write, exactly as it does today. The encryption protects the channel between participants, not what the underlying models observe.

Pricing

The free tier covers the 60-minute workspaces, sharing with another agent or person, and files up to 2 MB. A Personal plan at $7 per month adds persistent workspaces with durable history, raises file limits to 25 MB per file and 250 MB overall, and lets you keep any instant workspace. Invitees always join at no cost. There is no team plan yet; the site says the makers are collecting feedback on what teams would need.

Why it matters

Running several coding agents in parallel is becoming routine, but moving context between them remains awkward — copy-paste across terminals, scratch commits, messages to yourself. Jotbus proposes a generic, agent-agnostic answer, and because it speaks MCP, the set of compatible tools can grow well beyond the thirteen named at launch. The end-to-end encryption answers the obvious objection to routing work notes and logs through a third-party server, and the documented install footprint and uninstall path suggest the builders expect scrutiny.

Caveats remain: the provider-visibility note means the encryption covers storage and transit rather than the models themselves, and every security claim is currently the project's own. As an indicator of where agent tooling is heading — neutral shared infrastructure instead of single-vendor orchestration — it is a launch worth watching.