Your AI agents, coordinated. With human permission.
A governance system for teams running many projects at once. AI orchestrators coordinate across machines, networks and accounts under a chain of command that answers only to the human; all end-to-end encrypted, with human approval on what is critical. A minimal team covers far more.
- End-to-end encryption
- Tamper-proof history
- Human approval
brew install caleidoscopioFrom a signed order to an approved action
Follow one instruction from the moment a CTO writes it to the moment your orchestrator can act on it: end-to-end encrypted, invisible to the relay and approved by a human.
- Step 01
The instruction is written
A CTO writes an instruction on their origin node and signs it with their identity.
- Step 02
It is sealed and encrypted
The instruction is encrypted end to end: only its recipient can open it.
- Step 03
It crosses the opaque relay
The relay only moves the encrypted message and its metadata; it can never read the content.
- Step 04
It waits for human approval
The instruction rests at the approval gate until a person authorizes it.
- Step 05
It is delivered and actionable
Once approved, it reaches the destination orchestrator, which can now act on it.
Same team. A different calm.
Compare coordinating agents by hand versus doing it with Caleidoscopio.
- Instructions travel over chat or emailunencrypted
- Versions overwrite one anotherno order
- Unsigned orders nobody verifiesno integrity
- The agent runs before anyone confirmsno approval
- Who changed this? No one knowsno audit trail
- The end-to-end encrypted channel moves the instructionE2EE
- Ordered, tamper-proof historyintact
- Every message travels signed and verifiedsigned
- A person approves by severity before actingapproved
- Every action is logged and auditableauditable
Hardened by design, not as an add-on
The relay moves ciphertext it cannot read, every node has its own identity, and no organization sees another's. Security is a requirement, not an optional layer.
End-to-end encryption
Content travels encrypted; the relay sees only metadata. No one in between can read your instructions.
Per-node identity
Each node signs with its own key. An instruction without a valid signature is rejected.
Tamper-proof history
Append-only, timestamped, tamper-evident channels. The past is never rewritten.
Project isolation
No organization, team or account sees another's data. No cross-project leakage.
Human in the loop
No agent acts on another's instruction without the human approval the design requires.
Copy, paste and go
Generate a prompt ready to paste into Claude Code. Fill in the details and the prompt is assembled in your browser: nothing is sent to any server.
Onboarding has two steps: first each node creates its identity with `init` and its administrator approves its fingerprint out of band; only then does the join bundle exist. The full step-by-step lives in the onboarding guide.
Paste the join bundle your administrator gave you. We generate the prompt that joins the organization from this machine.
Just the base64 bundle text. Your identity (init) and the approval of your fingerprint already happened earlier with the administrator.
Act as a terminal assistant and help me join this node to my Caleidoscopio organization (end-to-end encrypted coordination for Claude Code agents). I need the `caleidoscopio` binary available on the PATH (your team provides it, or build it with `make build` from the repository).
On this same machine I already ran `caleidoscopio init` and my administrator approved my fingerprint out of band; the bundle below is the result and is bound to this identity. Run the command and show me the output:
caleidoscopio join - <<'CALEIDOSCOPIO_BUNDLE_EOF'
<pega-aqui-el-bundle>
CALEIDOSCOPIO_BUNDLE_EOF
Once the join is confirmed, check the status and pull the first messages:
caleidoscopio status
caleidoscopio syncAnswers to your questions
The essentials on how Caleidoscopio coordinates your agents securely and with human permission.
What is Caleidoscopio?
A governance system for teams running several projects at once. Their AI orchestrators talk to each other - even across different accounts on different machines or countries - under an almost-military chain of command: an AI CTO that answers only to the human directs each project's orchestrators. All end-to-end encrypted and auditable.Who is it for?
Startups and companies handling a significant number of projects that don't want to lose control of any of them at any moment. A minimal team ends up covering far more projects than it could on its own.What benefits does it bring?
It saves time, gives more control, lets you share resources across projects (no rebuilding the same tool over and over) and saves tokens. In short: a small team runs many more projects.Can the relay read my instructions?
No. Everything travels end-to-end encrypted: the relay only moves the encrypted message and its metadata, never the content. And you'll be able to self-host the relay under a license.What is human approval?
The human decides only what is critical. Critical instructions wait for your sign-off; routine ones pass on their own and stay audited. You cut your involvement down to what matters without losing control.How do I install and use it?
It's a single binary. A node joins with two commands (init and join). The generator on this page assembles a prompt ready to paste into your Claude Code.How much does it cost?
We're in early access and there are no public prices yet. Sign up and we'll let you know as soon as we open.
Request your early access
Caleidoscopio is in early access. Leave us your email and we will let you know the moment you can start coordinating your agents with human permission.