Zero-knowledge backup for AI agent state
Backup agents to the cloud, restore anywhere
Your agent’s memory, skills and context live on one machine — one crash and they’re gone. Klonna keeps an encrypted clone in the cloud: one command to save, one to restore anywhere.
npm i -g @klonna/cli- encrypted on-device · AES-256-GCM
- free plan
- macOS · Linux · CI
OpenClaw
online
Like git, for your agents.
Klonna doesn’t dump backups in a folder. Every project becomes an agent with its own timeline of snapshots, grouped into workspaces you can share — browse them on the web, restore them from the terminal.
/workspaceWorkspace — like an organization
Private to you, or shared with the team. Everyone inside browses and restores the same agents.
/agentAgent — like a repository
One per project, named for its directory — the full history of that agent’s snapshots in one place.
/backupBackup — like a commit
An encrypted snapshot of memory, skills and context at a moment in time, with a file index you can inspect before restoring.
Three commands. Total recall.
The whole workflow fits in muscle memory: save when it matters, see everything you’ve kept, and restore wherever you land next.
- 1
klonna saveSave
Run it in your agent’s project. Klonna detects Claude Code from .claude/ — or any agent with an AGENTS.md — archives the state, and encrypts it with AES-256-GCM before a single byte leaves your machine. Secrets like .env files and SSH keys are stripped by rule, every time.
- 2
klonna listBrowse
Every project is an agent — like a repo — holding its own stream of snapshots. List them in the terminal, or browse workspaces, agents and backups — most with a browsable file index — in the web dashboard.
- 3
klonna restoreRestore
New laptop, fresh VM, or a teammate’s machine: install, sign in, restore. The archive is verified and decrypted locally, and your agent picks up exactly where it left off.
Zero-knowledge, and we mean it.
Klonna is built so that we cannot read your agent’s memory — not as a policy, as a property of the math.
Encrypted before upload
AES-256-GCM, with the key derived from your passphrase via scrypt — all of it in the CLI, on your machine. The server only ever holds ciphertext it cannot decrypt. The optional file index (paths and sizes, never contents) and your note are the only readable metadata — and --no-manifest keeps the file index off the server entirely.
Your passphrase never travels
It is never sent, never stored, never recoverable by us. Zero-knowledge cuts both ways: we cannot read your backups, and we cannot reset your passphrase.
An offline recovery key, if you want one
klonna recovery setup mints an X25519 key you keep offline. Any backup saved afterward can also be unlocked with it — locally, like everything else. Set it up before your first save.
Secrets stay out by rule
A hard-coded exclusion list — .env files, private keys, .ssh/, .aws/, credentials.json, .netrc — holds even if you edit the manifest. Skipped files are printed in the output, never hidden.
Bring the team.
Open a shared workspace and the same agents land in everyone’s hands: invite teammates with a link, set admin or member roles, and anyone inside can browse and restore its backups. New laptop on day one? klonna restore, and the project’s agent shows up already trained.
Built for agents, too.
API keys make Klonna fully non-interactive for CI runners and headless machines. There’s even a machine-readable instructions file your coding agent can follow to install the CLI, sign in and back itself up — no human at the keyboard.
Questions, answered plainly.
Tap a question to expand it.
What exactly does Klonna back up?
Your agent’s file-based state. For Claude Code that’s .claude/ (memory, skills, settings), CLAUDE.md, .mcp.json and AGENTS.md. Any other agent that keeps its state in files works in generic mode with an AGENTS.md, and a .klonnarc.json manifest controls exactly which files are included.
Can you read my backups?
No. Each backup is encrypted on your machine with AES-256-GCM, under a key your passphrase derives via scrypt, before it is uploaded. The server only ever sees ciphertext; decryption happens only in the CLI, on your hardware. The one readable thing is the optional file index — paths and sizes, never contents — so the dashboard can show what a backup holds. Save with --no-manifest to omit it.
What happens if I lose my passphrase?
Without a recovery key, the backup is gone — zero-knowledge means we cannot reset it. So run klonna recovery setup before your first save: it mints an offline recovery key (klr_…) that can decrypt any backup made after that point if the passphrase is ever lost. Store it like a password.
Will my .env files or SSH keys end up in a backup?
No. The CLI enforces a hard-coded exclusion list — .env files, private keys, .ssh/, .aws/, credentials.json, .netrc and more — even if you edit the manifest. Excluded secrets are printed in the output, so nothing sensitive is silently skipped or silently included.
Can my team restore the same agents?
Yes. Create a shared workspace and invite teammates with a link. Members browse the workspace’s agents and restore any backup; owners and admins manage who is in. Encryption stays client-side — share the passphrase through your password manager, not through us.
Can my agent back itself up, or run Klonna in CI?
Yes. Create an API key and set KLONNA_API_KEY (or run klonna login --api-key) for non-interactive use on CI runners and headless machines. There is also a machine-readable instructions file your coding agent can follow to install, sign in and save on its own.
Which agents and platforms are supported?
Claude Code is detected automatically; any file-based agent works via generic mode with an AGENTS.md. The CLI runs anywhere Node 20.17+ does — macOS, Linux, CI runners, cloud devboxes — with a single npm install.
Never lose an agent again.
One command installs the CLI, one saves an encrypted backup — free to start, zero-knowledge by design. Set it up now, restore anywhere later.
npm i -g @klonna/cli